Extended ethernet fabric switches

<u>Field</u>

0001The present disclosure relates to network design. More specifically, the present disclosure relates to a method for a constructing a scalable switching system.

<u>Background</u>

0002The exponential growth of the Internet has made it a popular delivery medium for a variety of applications running on physical and virtual devices. Such applications have brought with them an increasing demand for bandwidth. As a result, equipment vendors race to build larger and faster switches with versatile capabilities. However, the size of a switch cannot grow infinitely. It is limited by physical space, power consumption, and design complexity, to name a few factors. Furthermore, switches with higher capability are usually more complex and expensive. More importantly, because an overly large and complex system often does not provide economy of scale, simply increasing the size and capability of a switch may prove economically unviable due to the increased per-port cost.

0003A flexible way to improve the scalability of a switch system is to build a fabric switch. A fabric switch is a collection of individual member switches. These member switches form a single, logical switch that can have an arbitrary number of ports and an arbitrary topology. As demands grow, customers can adopt a "pay as you grow" approach to scale up the capacity of the fabric switch.

0004Meanwhile, layer-2 (e.g., Ethernet) switching technologies continue to evolve. More routing-like functionalities, which have traditionally been the characteristics of layer-3 (e.g., Internet Protocol or IP) networks, are migrating into layer-2. Notably, the recent development of the Transparent Interconnection of Lots of Links (TRILL) protocol allows Ethernet switches to function more like routing devices. TRILL overcomes the inherent inefficiency of the conventional spanning tree protocol, which forces layer-2 switches to be coupled in a logical spanning-tree topology to avoid looping. TRILL allows routing bridges (RBridges) to be coupled in an arbitrary topology without the risk of looping by implementing routing functions in switches and including a hop count in the TRILL header.

0005While a fabric switch brings many desirable features to a network, some issues remain unsolved in efficiently facilitating extended fabric switches (e.g., across datacenters).

0006<patcit id="pcit0001" dnum="US2012158997A"><text>US 2012/158997</text></patcit> provides a network system and a method of address resolution. The network system includes a network, a plurality of virtual machines, a routing module and a path directory module. Each virtual machine includes an internet protocol (IP) address and N media access control (MAC) addresses, so as to connect the network through N transmission routes. The routing module detects and calculates states of the transmission routes. The path directory module receives and decodes an address resolution protocol (ARP) request presented by at least one source virtual machine to reply path information, which includes N MAC addresses corresponding to an IP address of a destination virtual machine and states of the N transmission routes. Thus, the virtual machines present the ARP request without broadcast, so that the problem of network congestion is solved.

0007"<nplcit id="ncit0001" npl-type="s"><text>Virtual eXtensible Local Area Network (VXLAN): A Framework for Overlaying Virtualized Layer 2 Networks over Layer 3 Networks" by Mahalingham et al, IETF draft October 16, 2013</text></nplcit>, describes Virtual eXtensible Local Area Networks (VXLAN), which is used to address the need for overlay networks within virtualized data centers accommodating multiple tenants. The scheme and the related protocols can be used in networks for cloud service providers and enterprise data centers.

0008"<nplcit id="ncit0002" npl-type="b"><text>Cloud computing networking: challenges and opportunities for innovations" by Azodolmolky, S., Wieder, P., and Yahyapour, R, Communications Magazine, IEEE, Volume 51, Issue 7, July 2013</text></nplcit>, discloses how cloud computing can materialize the vision of utility computing. Tenants can benefit from on-demand provisioning of compute, storage, and networking resources according to a pay-peruse business model. Tenants have only limited visibility and control over network resources. The owners of cloud computing facilities are also facing challenges in various aspects of providing and efficiently managing Infrastructure as a Service (IaaS) facilities.

SUMMARY

0009The present invention is defined in the appended claims.

0010A switch is provided which includes a fabric switch module and a learning module. The fabric switch module maintains a membership in a first fabric switch. A fabric switch includes a plurality of switches and operates as a single switch. The first fabric switch is in an extended fabric switch which further comprises a second fabric switch. The learning module identifies from a notification message from the second fabric switch a media access control (MAC) address learned at the second fabric switch. The learning module stores the MAC address in a local MAC table in association with an Internet Protocol (IP) address of the second fabric switch.

0011In a variation on this, the IP address is a floating IP address and associated with a plurality of member switches of the second fabric switch.

0012In a variation on this, the switch further comprises a discovery module which identifies a discovery message comprising the IP address as a source IP address. This discovery message is a multicast message. In response, the discovery module includes the IP address in a neighbor list of the extended fabric switch. This neighbor list includes a respective IP address associated with a respective fabric switch in the extended fabric switch.

0013In a variation on this, the switch further comprises a forwarding module which identifies the MAC address as destination MAC address of a packet encapsulated in a fabric encapsulation header. In response to identifying the MAC address, the forwarding module encapsulates the packet in a tunnel encapsulation header. The destination address of the tunnel encapsulation header is the IP address.

0014In a further variation, the source MAC address of the packet is associated with a first virtual machine running on a host machine coupled to the first fabric switch, and the destination MAC address of the packet is associated with a second virtual machine running on a host machine coupled to the second fabric switch. The first and second virtual machines are in the same virtual local area network (VLAN).

0015In a variation on this, the switch further comprises a forwarding module which encapsulates an Address Resolution Protocol (ARP) request in a tunnel encapsulation header. The destination address of the tunnel encapsulation header is the IP address.

0016In a variation on this, the switch further comprises a forwarding module which encapsulates a packet destined to a virtual machine in a tunnel encapsulation header. The destination address of the tunnel encapsulation header is the IP address. This virtual machine has been migrated from a host machine coupled to the first fabric switch to a host machine coupled to the second fabric switch.

0017In a variation on this, the switch further comprises a forwarding module which identifies a packet encapsulated in a first tunnel encapsulation header. The destination address of the tunnel encapsulation header corresponds to a hypervisor. The forwarding module further encapsulates the encapsulated packet in a second tunnel encapsulation header. The destination address of the second tunnel encapsulation header is the IP address.

0018In a variation on this, the switch further comprises a high availability module which operates the switch as an active forwarder of an active-standby high availability protocol for a virtual machine. This virtual machine has been migrated from a host machine coupled to the second fabric switch to a host machine coupled to the first fabric switch.

0019In a variation on this, the switch further comprises a forwarding module which identifies a packet encapsulated in a fabric encapsulation header. The egress switch identifier of the fabric encapsulation header corresponds to a member switch of the second fabric switch. The forwarding module further encapsulates the encapsulated packet in a tunnel encapsulation header. The destination address of the tunnel encapsulation header is the IP address.

0020In a further variation, the fabric encapsulation header is one or more of: (i) a Transparent Interconnection of Lots of Links (TRILL) header, wherein ingress and egress switch identifiers of the fabric encapsulation header are TRILL routing bridge (RBridge) identifiers; and (ii) an IP header, wherein source and destination addresses of the fabric encapsulation header are IP addresses.

0021Also disclosed herein are a method and a computer system having the same functionality as the switch described above. Moreover, the method and computer system share the same potential variations as the switch, as described above.

BRIEF DESCRIPTION OF THE FIGURES

0022<ul id="ul0001" list-style="none" compact="compact"><li><figref idref="f0001">FIG. 1</figref> illustrates an exemplary extended fabric switch, in accordance with an embodiment of the present invention.</li><li><figref idref="f0002">FIG. 2A</figref> presents a flowchart illustrating the process of an extension switch initiating a neighbor discovery in an extended fabric switch, in accordance with an embodiment of the present invention.</li><li><figref idref="f0003">FIG. 2B</figref> presents a flowchart illustrating the process of an extension switch responding to neighbor discovery in an extended fabric switch, in accordance with an embodiment of the present invention.</li><li><figref idref="f0004">FIG. 2C</figref> presents a flowchart illustrating the process of an extension switch establishing tunnels with neighbor fabric switches, in accordance with an embodiment of the present invention.</li><li><figref idref="f0005">FIG. 3A</figref> illustrates an exemplary forwarding of an Address Resolution Protocol (ARP) request in an extended fabric switch, in accordance with an embodiment of the present invention.</li><li><figref idref="f0006">FIG. 3B</figref> illustrates an exemplary forwarding of an ARP response among in an extended fabric switch, in accordance with an embodiment of the present invention.</li><li><figref idref="f0007">FIG. 4A</figref> presents a flowchart illustrating the process of an extension switch sharing learned media access control (MAC) addresses in an extended fabric switch, in accordance with an embodiment of the present invention.</li><li><figref idref="f0008">FIG. 4B</figref> presents a flowchart illustrating the process of an extension switch learning MAC addresses from a neighbor fabric switch, in accordance with an embodiment of the present invention.</li><li><figref idref="f0009">FIG. 5A</figref> illustrates an exemplary forwarding of a data packet in an extended fabric switch, in accordance with an embodiment of the present invention.</li><li><figref idref="f0010">FIG. 5B</figref> presents a flowchart illustrating the process of an extension switch forwarding a data packet to a neighbor fabric switch, in accordance with an embodiment of the present invention.</li><li><figref idref="f0011">FIG. 5C</figref> presents a flowchart illustrating the process of an extension switch receiving a data packet from a neighbor fabric switch, in accordance with an embodiment of the present invention.</li><li><figref idref="f0012">FIG. 6A</figref> illustrates exemplary coupling of virtual machines to an extended fabric switch, in accordance with an embodiment of the present invention.</li><li><figref idref="f0013">FIG. 6B</figref> illustrates an exemplary migration of a virtual machine across neighbor fabric switches, in accordance with an embodiment of the present invention.</li><li><figref idref="f0014">FIG. 7</figref> illustrates an exemplary Virtual Router Redundancy Protocol (VRRP) localization in an extended fabric switch, in accordance with an embodiment of the present invention.</li><li><figref idref="f0015">FIG. 8</figref> illustrates an exemplary work load management in an extended fabric switch, in accordance with an embodiment of the present invention.</li><li><figref idref="f0016">FIG. 9</figref> illustrates an exemplary direct forwarding of a data packet in an extended fabric switch, in accordance with an embodiment of the present invention.</li><li><figref idref="f0017">FIG. 10</figref> illustrates an exemplary switch in an extended fabric switch, in accordance with an embodiment of the present invention.</li></ul>

0023In the figures, like reference numerals refer to the same figure elements.

DETAILED DESCRIPTION

0024The following description is presented to enable any person skilled in the art to make and use the invention, and is provided in the context of a particular application and its requirements. Various modifications to the disclosed embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other embodiments and applications without departing from the spirit and scope of the present invention as defined by the appended claims. Thus, the present invention is not limited to the embodiments shown, but is to be accorded the widest scope consistent with the claims.

0025Furthermore, the present application describes various embodiments and implementations. The skilled person will understand that features from different embodiments and implementations may be combined as appropriate to create new embodiments and implementations of the present invention according to the particular circumstances.

<u>Overview</u>

0026In embodiments of the present invention, the problem of deploying a distributed fabric switch is solved by facilitating an extended fabric switch comprising a plurality of fabric switches, which can be in different datacenters. With existing technologies, a fabric switch typically is deployed within a datacenter and is not extended across multiple datacenters. As a result, virtual machine (VM) migration and workload management across datacenters may not be feasible. Moreover, when a member switch in a fabric switch learns media access control (MAC) address information (e.g., a learned MAC address and a corresponding virtual local area network (VLAN) tag), the switch shares that information only within the local fabric switch. Furthermore, when two fabric switches are interconnected, their connection is often established in layer-3 (e.g., via a wide area network (WAN)). However, such interconnection can be expensive since it relies on availability of layer-3 devices (e.g., routers) and may require extensive configuration of the layer-3 devices.

0027To solve this problem, a fabric switch is extended to one or more other fabric switches. These fabric switches, together, operate as an extended fabric switch. A respective fabric switch in an extended fabric switch can be referred to as a neighbor fabric switch of the extended fabric switch. Neighbor fabric switches of an extended fabric switch can be collocated in a local area network (LAN) or span a WAN (e.g., across multiple datacenters). In some embodiments, an extended fabric switch operates as a single fabric switch. A respective neighbor fabric switch includes one or more extension switches. These extension switches are capable of establishing tunnels with other extension switches in another neighbor fabric switch. These tunnels allow the neighbor fabric switches in an extended fabric switch to communicate with each other over a WAN.

0028During operation, an extension switch discovers other neighbor fabric switches. In some embodiments, the extension switch sends a discovery message to an All Switch Multicast (ASM) group to discover the neighbor fabric switches. This allows the extension switch to identify the neighbor fabric switch and extend the local fabric switch by establishing respective tunnels to the neighbor fabric switches. Furthermore, typically in a fabric switch, when a member switch learns a MAC address (and associated information, such as a VLAN tag), that member switch shares the learned MAC address (e.g., via a notification message) with other member switches. In an extended fabric switch, upon learning a MAC address (either locally or from another member switch), an extension switch shares the learned MAC address with other extension switches of neighbor fabric switches. As a result, a respective neighbor fabric switch can learn a MAC address learned in the extended fabric switch.

0029In a fabric switch, any number of switches coupled in an arbitrary topology may logically operate as a single switch. The fabric switch can be an Ethernet fabric switch or a virtual cluster switch (VCS), which can operate as a single Ethernet switch. Any member switch may join or leave the fabric switch in "plug-and-play" mode without any manual configuration. In some embodiments, a respective switch in the fabric switch is a Transparent Interconnection of Lots of Links (TRILL) routing bridge (RBridge). In some further embodiments, a respective switch in the fabric switch is an Internet Protocol (IP) routing-capable switch (e.g., an IP router). The TRILL protocol is described in Internet Engineering Task Force (IETF) Request for Comments (RFC) 6325, titled "Routing Bridges (RBridges): Base Protocol Specification," available at http://datatracker.ietf.org/doc/rfc6325/.

0030It should be noted that a fabric switch is not the same as conventional switch stacking. In switch stacking, multiple switches are interconnected at a common location (often within the same rack), based on a particular topology, and manually configured in a particular way. These stacked switches typically share a common address, e.g., an IP address, so they can be addressed as a single switch externally. Furthermore, switch stacking requires a significant amount of manual configuration of the ports and inter-switch links. The need for manual configuration prohibits switch stacking from being a viable option in building a large-scale switching system. The topology restriction imposed by switch stacking also limits the number of switches that can be stacked. This is because it is very difficult, if not impossible, to design a stack topology that allows the overall switch bandwidth to scale adequately with the number of switch units.

0031In contrast, a fabric switch can include an arbitrary number of switches with individual addresses, can be based on an arbitrary topology, and does not require extensive manual configuration. The switches can reside in the same location, or be distributed over different locations. These features overcome the inherent limitations of switch stacking and make it possible to build a large "switch farm," which can be treated as a single, logical switch. Due to the automatic configuration capabilities of the fabric switch, an individual physical switch can dynamically join or leave the fabric switch without disrupting services to the rest of the network.

0032Furthermore, the automatic and dynamic configurability of the fabric switch allows a network operator to build its switching system in a distributed and "pay-as-you-grow" fashion without sacrificing scalability. The fabric switch's ability to respond to changing network conditions makes it an ideal solution in a virtual computing environment, where network loads often change with time.

0033In this disclosure, the term "fabric switch" refers to a number of interconnected physical switches which form a single, scalable logical switch. These physical switches are referred to as member switches of the fabric switch. In a fabric switch, any number of switches can be connected in an arbitrary topology, and the entire group of switches functions together as one single, logical switch. This feature makes it possible to use many smaller, inexpensive switches to construct a large fabric switch, which can be viewed as a single logical switch externally. Although the present disclosure is presented using examples based on a fabric switch, embodiments of the present invention are not limited to a fabric switch. Embodiments of the present invention are relevant to any computing device that includes a plurality of devices operating as a single device.

0034The term "end device" can refer to any device external to a fabric switch. Examples of an end device include, but are not limited to, a host machine, a conventional layer-2 switch, a layer-3 router, or any other type of network device. Additionally, an end device can be coupled to other switches or hosts further away from a layer-2 or layer-3 network. An end device can also be an aggregation point for a number of network devices to enter the fabric switch. An end device can also host one or more virtual machines.

0035The term "switch" is used in a generic sense, and it can refer to any standalone or fabric switch operating in any network layer. "Switch" should not be interpreted as limiting embodiments of the present invention to layer-2 networks. Any device that can forward traffic to an external device or another switch can be referred to as a "switch." Any physical or virtual device (e.g., a virtual machine/switch operating on a computing device) that can forward traffic to an end device can be referred to as a "switch." Examples of a "switch" include, but are not limited to, a layer-2 switch, a layer-3 router, a TRILL RBridge, or a fabric switch comprising a plurality of similar or heterogeneous smaller physical and/or virtual switches.

0036The term "edge port" refers to a port on a fabric switch which exchanges data frames with a network device outside of the fabric switch (i.e., an edge port is not used for exchanging data frames with another member switch of a fabric switch). The term "inter-switch port" refers to a port which sends/receives data frames among member switches of a fabric switch. An edge port used in a tunnel between two neighbor fabric switches of a fabric switch can be referred to as an "extension port." The terms "interface" and "port" are used interchangeably.

0037The term "switch identifier" refers to a group of bits that can be used to identify a switch. Examples of a switch identifier include, but are not limited to, a media access control (MAC) address, an Internet Protocol (IP) address, and an RBridge identifier. Note that the TRILL standard uses "RBridge ID" (RBridge identifier) to denote a 48-bit intermediate-system-to-intermediate-system (IS-IS) System ID assigned to an RBridge, and "RBridge nickname" to denote a 16-bit value that serves as an abbreviation for the "RBridge ID." In this disclosure, "switch identifier" is used as a generic term, is not limited to any bit format, and can refer to any format that can identify a switch. The term "RBridge identifier" is also used in a generic sense, is not limited to any bit format, and can refer to "RBridge ID," "RBridge nickname," or any other format that can identify an RBridge.

0038The term "packet" refers to a group of bits that can be transported together across a network. "Packet" should not be interpreted as limiting embodiments of the present invention to layer-3 networks. "Packet" can be replaced by other terminologies referring to a group of bits, such as "message," "frame," "cell," or "datagram." The terms "packet" and "frame" are used interchangeably.

<u>Network Architecture</u>

0039<figref idref="f0001">FIG. 1</figref> illustrates an exemplary extended fabric switch, in accordance with an embodiment of the present invention. As illustrated in <figref idref="f0001">FIG. 1</figref>, fabric switches 120 and 130 is extended to each other to form an extended fabric switch 100. Fabric switch 120 includes member switches 122, 124, 126, and 128; and fabric switch 130 includes member switches 132, 134, 136, and 138. End devices 142 and 144 are coupled to fabric switch 120 via switches 126 and 128, respectively; and end devices 146 and 148 are coupled to fabric switch 130 via switches 136 and 138, respectively. A member switch, such as switch 126 or 136, which couples an end device via an edge port, can be referred to as an edge switch.

0040Fabric switches 120 and 130 are coupled to each other via a network 110. In this example, network 110 can be a WAN or the Internet. Fabric switches 120 and 130 can also be in a LAN. Switches 122 and 124 are coupled to layer-3 devices (e.g., routers) 112 and 114. Similarly, switches 126 and 128 are coupled to layer-3 devices 116 and 118. Switch 122 uses extension ports to couple to layer-3 devices 112 and 114. To send a packet to fabric switch 130, switch 122 sends the packet via one of the extension ports. Suppose that the packet is forwarded via network 110 to layer-3 device 116, which in turn, forwards the packet to switch 132. Switch 132 receives the packet via a local extension port. In this way, fabric switches 120 and 130 communicate via network 110. It should be noted that devices in network 110 and extended fabric switch 100 can be physical or virtual.

0041In some embodiments, fabric switches 120 and 130 internally operate as respective TRILL networks (e.g., forward data packet based on the TRILL protocol). A respective member switch of fabric switches 120 and 130 can then be a TRILL RBridge (e.g., has an RBridge identifier which identifies a member switch in the corresponding fabric switch). In some further embodiments, fabric switches 120 and 130 internally operate as respective IP networks (e.g., forward data packet based on the IP protocol). A respective member switch of fabric switches 120 and 130 can then be an IP-capable switch (e.g., has an IP address which identifies a member switch in the corresponding fabric switch and/or a larger network). An IP-capable switch can calculate and maintain a local IP routing table (e.g., a routing information base or RIB), and is capable of forwarding packets based on its IP addresses.

0042Switches in a fabric switch use edge ports to communicate with end devices (e.g., non-member switches) and inter-switch ports to communicate with other member switches. Data communication via an edge port can be based on Ethernet and via an inter-switch port can be based on IP and/or TRILL protocol. For example, switch 126 of fabric switch 120 is coupled to end device 142 via an edge port and to switches 122 and 128 via inter-switch ports. Switch 126 can communicate with end device 142 based on Ethernet and with switch 122 and 128 based on IP or TRILL protocol. It should be noted that control message exchange via inter-switch ports can be based on a different protocol (e.g., Internet Protocol (IP) or Fibre Channel (FC) protocol).

0043With existing technologies, fabric switch 120 and 130 typically are deployed within respective datacenter and are not extended across multiple datacenters. As a result, VM migration and workload management across the respective datacenters of fabric switches 120 and 130 may not be feasible. When a member switch, such as switch 126, learns MAC address information (e.g., a learned MAC address and a corresponding virtual local area network (VLAN) tag), switch 126 shares the learned information only with other member switches of fabric switch 120 (i.e., switches 122, 124, and 128). Furthermore, when fabric switches 120 and 130 are interconnected, their connection is often established in layer-3 devices in network 110. However, such layer-3 devices are outside of fabric switches 120 and 130. As a result, such connection relies on availability of layer-3 devices in network 110. Moreover, to facilitate interconnection between fabric switches 120 and 130, extensive configuration of layre-3 devices 112, 114, 116, and 118 may be required.

0044To solve this problem, fabric switch 120 is extended to fabric switch 130, and fabric switch 130 is extended to fabric switch 120. Fabric switches 120 and 130, together, operate as extended fabric switch 100. Fabric switch 120 and 130 can be referred to as neighbor fabric switches of extended fabric switch 100. Neighbor fabric switches 120 and 130 can be collocated in a LAN or span a WAN, such as network 110. Neighbor fabric switches 120 and 130 can also be in different datacenters. Extended fabric switch 100 can operate as a single fabric switch. Switches 122 and 124 are extension switches of fabric switch 120, and switches 132 and 134 are extension switches of fabric switch 130. These extension switches are capable of establishing tunnels with other extension switches in another neighbor fabric switch. These tunnels allow neighbor fabric switches 120 and 130 to operate as a single extended fabric switch 100.

0045During operation, extension switch 122 discovers other neighbor fabric switches. In some embodiments, a respective extension switch in extended fabric switch 100 is configured with a membership to an All Switch Multicast (ASM) group. This ASM group can be used to exchange control messages between neighbor fabric switches 120 and 130. During the neighbor discovery process, extension switch 122 sends a discovery message, which is a multicast message, to the ASM group to discover neighbor fabric switch 130. Extension switches 124, 132, and 134 receive the discovery message. In some embodiments, since extension switch 124 is in the same fabric switch 120, extension switch 124 discards the discovery message.

0046Extension switches 132 and 134 receive the discovery message and discover fabric switch 120. The discovery message can include a fabric IP address associated with fabric switch 120. This fabric IP address can be a floating IP address and a respective extension switch of fabric switch 120 can terminate forwarding of a packet with the fabric IP address as the destination address. In some embodiments, the fabric IP address is assigned to a logical interface of extension switches 122 and/or 124. This logical interface can correspond to one or more physical ports in extension switches 122 and/or 124.

0047Upon receiving the discovery message, extension switches 132 and 134 discover fabric switch 120 to be a neighbor of their local fabric switch 130 and add the fabric IP address of fabric switch 120 to their neighbor list. In the same way, extension switch 124 also sends a discovery packet to the ASM group. On the other hand, extension switches 122 and 124 discover fabric switch 130 to be a neighbor of their local fabric switch 120 and add the fabric IP address of fabric switch 130 to their neighbor list. Once neighbor discovery is completed and a respective extension switch learns a respective fabric IP address of a corresponding neighbor fabric switch, extension switches 122, 124, 132, and 134 establish a tunnel mesh 102. Tunnel mesh 102 includes a full mesh of tunnels between a respective fabric switch pair in extended fabric switch 100. Examples of tunnels in tunnel mesh 102 include, but are not limited to, Virtual Extensible LAN (VXLAN) tunnel, Generic Routing Encapsulation (GRE) tunnel, and Network Virtualization using GRE (NVGRE) tunnel.

0048In some embodiments, extended fabric switch 100 includes a directory server. One of the extension switches in extended fabric switch 100 can operate as the directory server. A respective other extension switch in extended fabric switch 100 can be configured with the address of the directory server. The directory server maintains a list of fabric IP addresses, a respective of which is associated with a corresponding neighbor fabric switch. During operation, these other extension switches query the directory server to obtain the list of IP addresses. The directory server can send a response message responding to a respective query comprising the list of IP addresses. Another extension switch in extended fabric switch 100 can operate as a standby directory server, which can be in a different neighbor fabric switch. For example, switch 122 can operate as the directory server and switch 132 can operate as the standby directory server.

0049Furthermore, suppose that switch 126 learns the MAC address of end device 142. Switch 126 generates a notification message comprising the learned MAC address (e.g., in the payload) and sends the notification message to a respective other member switch of fabric switch 120 (i.e., switches 122, 124, and 128). Upon receiving the notification message, switch 122 creates an entry in the local MAC table (typically stored in a Content-Addressable Memory (CAM)) comprising the learned MAC address (and associated information) in association with the switch identifier. Furthermore, switch 122 creates a notification message comprising the learned MAC address (and associated information) and sends the notification message to the ASM group.

0050This notification message has the fabric IP address of fabric switch 120 as the source address and the multicast IP address of ASM group as the destination address. The notification message is forwarded via network 110 to reach extension switches 132 and 134. Upon receiving the notification message, extension switches 132 and 134 extract the MAC address (and associated information) and store the extracted information in association with the source IP address of the notification message.

0051In some embodiments, tunnel mesh 102 is associated with a tunnel identifier. Extension switches in extended fabric switch 100 use this tunnel identifier to identify the tunnel used to extend a fabric switch (i.e., tunnel mesh 102). In some embodiments, the tunnel identifier is a VXLAN Network Identifier (VNI). If a tunnel is not associated with the tunnel identifier, it is not in tunnel mesh 102 and the notification message comprising the learned MAC address is not forwarded via that tunnel. This allows fabric switches 120 and 130 to have other tunnel(s) in addition to tunnel mesh 102. In some embodiments, a respective extension switch in extended fabric switch 100 is preconfigured with the fabric IP addresses of fabric switch 120 and 130 and the tunnel identifier of tunnel mesh 102. Additionally, the extension switch can also be preconfigured with a VLAN tag for VXLAN encapsulation and a list of VLAN tags which are extended in both fabric switches 120 and 130.

0052In some embodiments, multiple virtual fabric switches can deployed within an extended fabric switch. A respective virtual fabric switch is associated with a separate tunnel mesh (e.g., with a different tunnel identifier). In some embodiments, openFlow can be used to define the virtual fabric switches. To improve the performance of Address Resolution Protocol (ARP) processing, extension switches can maintain an ARP proxy to limit ARP requests between neighbor fabric switches. Furthermore, to efficiently manage learned MAC addresses, an extension switch can learn the MAC address from a neighbor fabric switch if the extension switch is in communication (e.g., forwards frame) with that MAC address. Moreover, multicast forwarding can be optimized by forwarding multicast packet only to extension switches with a listener.

<u>Neighbor Discovery</u>

0053<figref idref="f0002">FIG. 2A</figref> presents a flowchart illustrating the process of an extension switch initiating neighbor discovery in an extended fabric switch, in accordance with an embodiment of the present invention. During operation, the extension switch obtains the multicast IP address of the ASM group for forwarding control messages (operation 202). The extension switch can send a join message to join the ASM group, or be preconfigured with a membership to the ASM group. The extension switch then generates a layer-3 discovery message (e.g., an IP multicast packet) (operation 204). The extension switch sets the local fabric IP address as the source address of the layer-3 header (operation 208). The extension switch sets the multicast IP address of the ASM group as the destination address of the layer-3 header (operation 210). The extension switch then identifies the extension port(s) associated with the multicast IP address (operation 212) and transmits the discovery message via the identified port(s) (operation 214).

0054<figref idref="f0003">FIG. 2B</figref> presents a flowchart illustrating the process of an extension switch responding to neighbor discovery in an extended fabric switch, in accordance with an embodiment of the present invention. During operation, the extension switch receives a layer-3 discovery message via a local extension port (operation 232). The extension switch obtains the source IP address of the layer-3 header (operation 234). This discovery message indicates that a neighbor fabric switch is establishing neighbor adjacency. The extension switch thus discovers a neighbor fabric switch based on the discovery message (operation 236) and includes the obtained IP address in the neighbor list (operation 238).

0055<figref idref="f0004">FIG. 2C</figref> presents a flowchart illustrating the process of an extension switch establishing tunnels with a neighbor fabric switch, in accordance with an embodiment of the present invention. During operation, the extension switch obtains an IP address from a local neighbor list (operation 252) and establishes a tunnel using the local and obtained fabric IP addresses as tunnel endpoints (operation 254). This tunnel allows an extension switch to terminate tunnel-encapsulated packets. The extension switch assigns a tunnel identifier (e.g., a VNI) to the established tunnel and stores a mapping between the tunnel identifier and the tunnel (e.g., the fabric IP address) (operation 256). The extension switch determines the outer VLAN tag for the tunnel (e.g., an outer VLAN tag of a VXLAN header) and stores a mapping between the outer VLAN tag and the tunnel (operation 258). The extension switch also determines and stores the VLAN(s) that are extended via the tunnel (operation 260). It should be noted that these mappings can be preconfigured in the extension switch.

<u>ARP Forwarding</u>

0056<figref idref="f0005">FIG. 3A</figref> illustrates an exemplary forwarding of an ARP request in an extended fabric switch, in accordance with an embodiment of the present invention. During operation, end device 142 initiates a communication with end device 148 and sends an ARP request 302 (which can also be referred to as ARP query 302) comprising the IP address of end device 148. Edge switch 126 receives ARP request 302 via an edge port. Switch 126 learns the MAC address of end device 142 and adds the MAC address to its local MAC table (can also be referred to as a forwarding table) in association with the edge port (e.g., based on a port identifier). Switch 126 also generates a notification message comprising the learned MAC address and sends the notification message to switches 122, 124, and 128. In turn, switches 122, 124, and 128 learn the MAC address of end device 142 and add the MAC address to their respective local MAC tables in association with switch identifier 304 (e.g., an RBridge identifier or an IP address) of switch 126. In some embodiments, switches 122, 124, and 128 further associate the MAC address of end device 142 with the edge port of switch 126 (e.g., based on a port identifier).

0057Switch 126 encapsulates ARP request 302 with a fabric encapsulation (e.g., TRILL or IP encapsulation) to create fabric-encapsulated packet 322 (operation 332). Since switch 126 does not know the destination (i.e., has not learned the destination MAC address), switch 126 assigns an "all-switch" switch identifier 306 as the egress switch identifier and switch identifier 304 of switch 126 as the ingress switch identifier of the encapsulation header. Switch 126 forwards packet 322 to a respective switch in fabric switch 120. It should be noted that forwarding includes determining an egress (or output) port associated with the destination address and transmitting via the determined egress port.

0058When packet 322 reaches extension switch 122, switch 122 decapsulates packet 322 to extract ARP request 302 (operation 334). Since switch 122 has already discovered neighbor fabric switch 130, switch 122 encapsulates ARP request 302 with a tunnel encapsulation corresponding to tunnel mesh 102 to generate tunnel-encapsulated packet 324 (operation 336). It should be noted that an encapsulation (e.g., a tunnel or fabric encapsulation) includes encapsulating the packet (e.g., ARP request 302) in an encapsulation header corresponding to the encapsulation. Switch 122 assigns fabric IP address 310 of fabric switch 130 as the destination address and fabric IP address 308 of fabric switch 120 as the source address of the encapsulation header. Switch 122 forwards packet 324 to fabric switch 130 via layer-3 device 112 or 114, as described in conjunction with <figref idref="f0001">FIG. 1</figref>.

0059It should be noted that upon learning the MAC address of end device 142, switch 122 can send a notification message to switches 132 and 134. Upon receiving the notification message, switches 132 and 134 stores the MAC address of end device 142 in association with fabric IP address 308. Since fabric IP address 310 is a floating IP address, switch 132 or 134 can terminate forwarding of packet 324. Suppose that switch 132 receives packet 324, terminates forwarding, and decapsulates the tunnel encapsulation to extract ARP request 302 (operation 338). Switch 132 can also learn the MAC address of end device 142 from ARP request 302. Switch 132 can also determine whether any local end device (i.e., coupled with switch 132) corresponds to the IP address in ARP request 302.

0060Switch 132 then encapsulates ARP request 302 with a fabric encapsulation to create fabric-encapsulated packet 326 (operation 340). Since switch 132 does not know the destination (i.e., has not learned the destination MAC address), switch 132 assigns an "all-switch" switch identifier 314 (which can be the same as identifier 306) as the egress switch identifier and switch identifier 312 of switch 132 as the ingress switch identifier of the encapsulation header. Switch 132 forwards packet 326 to a respective switch in fabric switch 130. Upon receiving packet 326, switch 138 decapsulates the fabric encapsulation to extract ARP request 302 (operation 342). Switch 138 learns the MAC address of end device 142 and stores the learned MAC address in association with switch identifier 312 of switch 132. Switch 138 can flood its local edge ports with ARP request 302. End device 148 thus receives ARP request 302 and learns MAC address of end device 142. End device 148 determines that the IP address in ARP request 302 is assigned to end device 148 and generates an ARP response comprising the MAC address of end device 148.

0061<figref idref="f0006">FIG. 3B</figref> illustrates an exemplary forwarding of an ARP response among in an extended fabric switch, in accordance with an embodiment of the present invention. During operation, end device 148 generates and sends an ARP response 352 responding to ARP request 302. The destination MAC address of ARP response 352 corresponds to end device 142. Edge switch 138 receives ARP response 352 via an edge port. Switch 138 learns the MAC address of end device 148 and adds the MAC address to its local MAC table in association with the edge port. Switch 138 also generates a notification message comprising the learned MAC address and sends the notification message to switches 132, 134, and 136. In turn, switches 132, 134, and 138 learn the MAC address of end device 148 and add the MAC address to their respective local MAC tables in association with switch identifier 354 (e.g., an RBridge identifier or an IP address) of switch 138. In some embodiments, switches 132, 134, and 136 further associate the MAC address of end device 148 with the edge port of switch 138.

0062Switch 138 encapsulates ARP response 352 with a fabric encapsulation (e.g., TRILL or IP encapsulation) to create fabric-encapsulated packet 372 (operation 382). Since switch 138 already knows the destination MAC address, which is associated with switch identifier 312 of switch 132, switch 138 assigns switch identifier 312 as the egress switch identifier and switch identifier 354 of switch 138 as the ingress switch identifier of the encapsulation header. Switch 138 forwards packet 372 to switch 132. When packet 372 reaches switch 132, switch 132 decapsulates packet 372 to extract ARP response 352 (operation 384). Since switch 132 has stored the MAC address of end device 142 in association with fabric IP address 308, switch 122 encapsulates ARP response 352 with a tunnel encapsulation corresponding to tunnel mesh 102 to generate tunnel-encapsulated packet 374 (operation 386). Switch 132 assigns fabric IP address 308 as the destination address and fabric IP address 310 as the source address of the encapsulation header. Switch 132 forwards packet 374 to fabric switch 120 via layer-3 device 116 or 118, as described in conjunction with <figref idref="f0001">FIG. 1</figref>.

0063Since fabric IP address 308 is a floating IP address, switch 122 or 124 can terminate forwarding of packet 374. Suppose that switch 122 receives packet 374, terminates forwarding, and decapsulates the tunnel encapsulation to extract ARP response 352 (operation 388). Switch 122 then encapsulates ARP response 352 with a fabric encapsulation to create fabric-encapsulated packet 376 (operation 390). Since switch 122 stores the MAC address in association with switch identifier 304, switch 122 assigns switch identifier 304 as the egress switch identifier and switch identifier 356 of switch 122 as the ingress switch identifier of the encapsulation header. Switch 122 forwards packet 376 to switch 126 based on switch identifier 304. Upon receiving packet 376, switch 126 decapsulates the fabric encapsulation to extract ARP response 352 (operation 392). Switch 126 forwards ARP response 352 via the edge port coupling end device 148. End device 148 thus receives ARP response 352 and learns MAC address of end device 148.

<u>MAC Sharing in Extended Fabric Switch</u>

0064<figref idref="f0007">FIG. 4A</figref> presents a flowchart illustrating the process of an extension switch sharing learned MAC addresses in an extended fabric switch, in accordance with an embodiment of the present invention. During operation, the extension switch receives a fabric notification message comprising one or more MAC address(es) (and associated information, such as corresponding VLAN tags) learned at a remote switch of the local fabric switch (operation 402). In some embodiments, the fabric notification message is based on an internal messaging service of the fabric switch. The extension switch extracts the MAC address(es) (and associated information) from the notification message (operation 404) and obtains the ingress switch identifier of the fabric notification message (operation 406). The extension switch updates the local MAC table with the extracted MAC address(es) (and associated information) in association with the obtained switch identifier (operation 408).

0065The extension switch also maps the local fabric IP address to extracted MAC address(es) (and associated information) (operation 410) and generate a notification message comprising the mapping based on a tunnel encapsulation (e.g., VXLAN encapsulation) (operation 412). The extension switch sets the local fabric IP address as the source address of the encapsulation header (e.g., VXLAN header) (operation 414). The extension switch sets the multicast IP address of the ASM group as the destination address of the encapsulation header (operation 416). The extension switch then identifies the extension port(s) associated with the multicast IP address (operation 418) and transmits the notification message via the identified port(s) (operation 420).

0066<figref idref="f0008">FIG. 4B</figref> presents a flowchart illustrating the process of an extension switch learning MAC addresses from a neighbor fabric switch, in accordance with an embodiment of the present invention. During operation, the extension switch receives a tunnel-encapsulated notification message (operation 452) and decapsulates the tunnel encapsulation to extract the notification message (operation 454). The extension switch obtains a mapping between a fabric IP address of a remote fabric switch and MAC address(es) (and associated information) learned at the remote fabric switch from the notification message (operation 456). The extension switch updates its local MAC table with the obtained mapping (operation 458).

<u>Data Forwarding</u>

0067<figref idref="f0009">FIG. 5A</figref> illustrates an exemplary forwarding of a data packet in an extended fabric switch, in accordance with an embodiment of the present invention. During operation, end device 142 initiates a communication with end device 148 and sends an Ethernet frame 502 destined to end device 148. Edge switch 126 receives Ethernet frame 502 via an edge port. Suppose that switch 126 has already learned and stored the MAC address of end device 148 in association with switch 122, as described in conjunction with <figref idref="f0005">FIG. 3A</figref>. Switch 126 encapsulates Ethernet frame 502 with a fabric encapsulation (e.g., TRILL or IP encapsulation) to create fabric-encapsulated packet 522 (operation 532). Since switch 126 knows the destination MAC address, switch 126 assigns switch identifier 356 of switch 122 as the egress switch identifier and switch identifier 304 of switch 126 as the ingress switch identifier of the encapsulation header. Switch 126 forwards packet 522 to switch 122 based on switch identifier 356.

0068When packet 522 reaches extension switch 122, switch 122 decapsulates packet 522 to extract Ethernet frame 502 (operation 534). Since switch 122 has already discovered neighbor fabric switch 130, switch 122 encapsulates Ethernet frame 502 with a tunnel encapsulation corresponding to tunnel mesh 102 to generate tunnel-encapsulated packet 524 (operation 536). Switch 122 assigns fabric IP address 310 of fabric switch 130 as the destination address and fabric IP address 308 of fabric switch 120 as the source address of the encapsulation header. Switch 122 forwards packet 524 to fabric switch 130 via layer-3 device 112 or 114, as described in conjunction with <figref idref="f0001">FIG. 1</figref>.

0069Since fabric IP address 310 is a floating IP address, switch 132 or 134 can terminate forwarding of packet 524. Suppose that switch 132 receives packet 524, terminates forwarding, and decapsulates the tunnel encapsulation to extract Ethernet frame 502 (operation 538). Switch 132 then encapsulates Ethernet frame 502 with a fabric encapsulation to create fabric-encapsulated packet 526 (operation 540). Since switch 132 knows the destination MAC address, switch 132 assigns switch identifier 354 of switch 138 as the egress switch identifier and switch identifier 312 of switch 132 as the ingress switch identifier of the encapsulation header. Switch 132 forwards packet 526 to switch 138 based on switch identifier 354. Upon receiving packet 526, switch 138 decapsulates the fabric encapsulation to extract Ethernet frame 502 (operation 542). Switch 126 forwards Ethernet frame 502 via the edge port coupling end device 148. End device 148 thus receives Ethernet frame 502 sent from end device 142.

0070<figref idref="f0010">FIG. 5B</figref> presents a flowchart illustrating the process of an extension switch forwarding a data packet to a neighbor fabric switch, in accordance with an embodiment of the present invention. During operation, the extension switch receives a fabric-encapsulated packet via a local inter-switch port (operation 552). The extension switch checks whether the local switch is the egress switch for the packet (operation 554). If the local switch is not the egress switch, the extension switch identifies an egress inter-switch port for the packet based on the egress switch identifier and forwards the fabric-encapsulated packet via the identified port (operation 556). Otherwise, the extension switch decapsulates the received packet to extract the inner packet (e.g., an Ethernet frame) and obtains the destination MAC address of the inner packet (operation 558).

0071The extension switch then checks whether the destination MAC address is a locally learned MAC address (operation 560). If the destination MAC address is a locally learned MAC address, the extension switch identifies an egress edge port based on the destination MAC address (e.g., from a local MAC table) and forwards the inner packet via the identified port (operation 562). Otherwise, the extension switch obtains the IP address mapped to the destination MAC address from the local MAC table (operation 564). In some embodiments, the destination MAC address is mapped to a tunnel identifier, which is associated with the IP address. The MAC address can be further mapped with a VLAN tag in the local MAC table. The extension switch encapsulates the inner packet in a tunnel encapsulation (e.g., VXLAN encapsulation) (operation 566), as described in conjunction with <figref idref="f0009">FIG. 5A</figref>.

0072The extension switch sets the local fabric IP address as the source address of the encapsulation header (e.g., a VXLAN header) (operation 568). The extension switch sets the obtained IP address as the destination address of the encapsulation header (operation 570). The extension switch then identifies the extension port associated with the obtained IP address (operation 572) and transmits the tunnel-encapsulated packet via the identified port (operation 574).

0073<figref idref="f0011">FIG. 5C</figref> presents a flowchart illustrating the process of an extension switch receiving a data packet from a neighbor fabric switch, in accordance with an embodiment of the present invention. During operation, the extension switch receives a tunnel-encapsulated packet via a local extension port (operation 576) and checks whether the local switch is the tunnel endpoint (operation 578). In some embodiments, the local switch is the tunnel endpoint if the destination address of the encapsulation header is a local fabric IP address. If the local switch is the tunnel endpoint, the extension switch decapsulates the received packet to extract the inner packet and obtains the destination MAC address of the inner packet (operation 580).

0074The extension switch then checks whether the destination MAC address is a locally learned MAC address (operation 582). If the destination MAC address is a locally learned MAC address, the extension switch identifies an egress edge port based on the destination MAC address (e.g., from a local MAC table) and forwards the inner packet via the identified port (operation 584). In some embodiments, the egress edge port is further selected based on a VLAN tag. If the destination MAC address is not a locally learned MAC address, the extension switch obtains a switch identifier mapped to the destination MAC address from the local MAC table (operation 586). The switch identifier can be further mapped to a VLAN tag.

0075The extension switch encapsulates the inner packet with a fabric encapsulation (e.g., TRILL encapsulation) (operation 588), as described in conjunction with <figref idref="f0009">FIG. 5A</figref>. The extension switch sets the local switch identifier as the ingress switch identifier of the encapsulation header (e.g., a TRILL header) (operation 590). The extension switch sets the obtained switch identifier as the egress switch identifier of the encapsulation header (operation 592). The extension switch then identifies the inter-switch port associated with the obtained switch identifier (operation 594) and transmits the fabric-encapsulated packet via the identified port (operation 596).

<u>Virtual Machine Management</u>

0076<figref idref="f0012">FIG. 6A</figref> illustrates exemplary coupling of virtual machines to an extended fabric switch, in accordance with an embodiment of the present invention. Extended fabric switch 100 can couple virtual machines in one or more datacenters. To do so, extended fabric switch 100 facilitates communication and migration for virtual machines in different datacenters. In this example, host machine (which is an end device) 610 is coupled to switch 126. Host machine 610 hosts a hypervisor 612 which runs virtual machines 614, 616, and 618 on host machine 610. Similarly, host machine 620 is coupled to switch 138. Host machine 620 hosts a hypervisor 622 which runs virtual machines 626 and 628 on host machine 620.

0077In some embodiments, a respective member switch of extended fabric switch 100 can facilitate tunneling (i.e., can operate as a tunnel endpoint and forward tunnel-encapsulated packets). Suppose that hypervisors 612 and 622 can also operate as tunnel endpoints. During operation, virtual machine 614 initiates communication with virtual machine 626 and sends a packet to hypervisor 612. Hypervisors 612 and 622 can be configured with each other's IP addresses and learn each other's MAC addresses based on ARP resolution, as described in conjunction with <figref idref="f0005">FIGs. 3A</figref> and <figref idref="f0006">3B</figref>. To enable communication between virtual machines 614 and 626, hypervisor 612 establishes a tunnel 602 with hypervisor 622 via extension switches 122 and 134. Hypervisor 612 encapsulates the packet from virtual machine 614 with a tunnel encapsulation corresponding to tunnel mesh 102. The source and destination addresses of this encapsulation header are IP addresses of hypervisors 612 and 614, respectively.

0078Hypervisor 612 assigns the MAC address of switch 122 as the next hop MAC address of the tunnel encapsulation and sends the packet. The packet is received by switch 122. Switch 122 further encapsulates the packet in another tunnel encapsulation corresponding to tunnel mesh 102. The source and destination addresses of this encapsulation header is the fabric IP addresses of fabric switches 120 and 130, respectively. Switch 122 forwards the encapsulated packet via network 110. Switch 132 (or 134) receives the packet and decapsulates the outer tunnel encapsulation. Switch 132 then looks up the destination IP address of the inner encapsulation and forwards the tunnel-encapsulated packet to hypervisor 622. Upon receiving the packet, hypervisor 620 decapsulates the tunnel encapsulation and delivers the inner packet to virtual machine 626.

0079Suppose that hypervisors 612 and 622 do not operate as tunnel endpoints. Under such a scenario, virtual machine 614 sends a packet (e.g., an Ethernet frame) to hypervisor 612. Hypervisor 612 receives the packet and sends the packet to switch 126, which in turn, encapsulates the packet with a fabric encapsulation and forwards the fabric-encapsulated packet to switch 122. Switch 122 decapsulates the fabric encapsulation, encapsulates the packet with a tunnel encapsulation, and forwards the fabric-encapsulated packet via network 110. Switch 132 receives the packet, decapsulates the tunnel encapsulation, encapsulates the packet in fabric encapsulation, and forwards via fabric switch 130, as described in conjunction with <figref idref="f0009">FIG. 5A</figref> Switch 138 receives the fabric-encapsulated packet, decapsulates the fabric encapsulation, and forwards via a local edge port which couples host machine 620. Hypervisor 622 receives the packet and delivers the packet to virtual machine 626.

0080<figref idref="f0013">FIG. 6B</figref> illustrates an exemplary migration of a virtual machine across neighbor fabric switches, in accordance with an embodiment of the present invention. In the example in <figref idref="f0013">FIG. 6B</figref>, VCenter 632 manages virtual machine migrations for the datacenter comprising fabric switch 120. During operation, virtual machine 614 is in communication (e.g., has an ongoing Transport Control Protocol (TCP) connection) with a user end device 652. Suppose that virtual machine 614 migrates to host machine 620 (denoted with dashed lines). In some embodiments, VCenter 632 manages the migration and notifies a Global Server Load Balancing (GSLB) controller 642. GSLB controller 642 balances load for virtual machines on host machines 610 and 620. GLSB controller 642 updates a Domain Name Service (DNS) server 644 regarding the migration and its associated updates (e.g., a new IP address in the IP subnet of host machine 620).

0081In some embodiments, extension switches of extended fabric switch 100 are aware of the migrated virtual machines (e.g., based on a notification from VCenter 632). Since end device 652 is in communication with virtual machine 614, end device 652 continues to forward packets to fabric switch 120. Extension switch 122 (or 124) receives such a packet, encapsulates it with a tunnel encapsulation corresponding to tunnel mesh 102, and forwards the tunnel-encapsulated packet to fabric IP address 310 of fabric switch 130. Switch 132 (or 134) receives the packet, identifies fabric IP address 310 as the destination address of the encapsulation header, and decapsulates the tunnel encapsulation.

0082Switch 132 encapsulates the packet with a fabric encapsulation and forwards the fabric-encapsulated packet to switch 138, which, in turn, decapsulates the fabric encapsulation and forwards the packet to hypervisor 622 in host machine 620. Hypervisor 622 delivers the packet to virtual machine 614. In this way, even when virtual machine 614 migrates across datacenters, existing connections can continue to operate. On the other hand, when a user end device 654 initiates a new connection and queries DNS server 644 for the location of virtual machine 614, the response from DNS server 644 points to the new location (e.g., a new IP address) of virtual machine 614. As a result, end device 654 sends packets to fabric switch 130 via layer-3 device 132 or 134.

0083In some embodiments, a respective member switch of a fabric switch includes a set of port profiles. A port profile includes configurations of a port. Examples of such configuration include, but are not limited to, Quality of Service (QoS) configuration, VLAN configuration, security configuration, and network configuration (e.g., FC configuration). A port profile is associated with one or more MAC addresses. When a switch identifies a MAC address as a source address in an Ethernet header of an Ethernet frame, the switch applies the corresponding port profile to the ingress port of the Ethernet frame. For example, suppose that a port profile is associated with the MAC addresses of virtual machines 614 and 616. Upon receiving an Ethernet frame from virtual machine 616 via port 662, switch 126 identifies the MAC address of virtual machine 616 as the source MAC address of the Ethernet header and applies the corresponding port profile to port 662.

0084In some embodiments, the port profiles are synchronized in extended fabric switch 100. For example, extension switches 122 and 124 synchronize the port profiles of fabric switch 120 with extension switches 132 and 134. Similarly, extension switches 132 and 134 synchronize the port profiles of fabric switch 130 with extension switches 122 and 124. In this way, the port profiles of fabric switches 120 and 130 are available in both fabric switches. After virtual machine 614 migrates to host machine 620, switch 138 receives an Ethernet frame from virtual machine 614 via port 664. Since the port profiles are synchronized in extended fabric switch 100, the port profile associated with the MAC address of virtual machine 614 is available in the member switches of fabric switch 130. Switch 138 identifies the MAC address of virtual machine 614 as the source MAC address of the Ethernet header and applies the corresponding port profile to port 664.

<u>VRRP Localization</u>

0085<figref idref="f0014">FIG. 7</figref> illustrates an exemplary VRRP localization in an extended fabric switch, in accordance with an embodiment of the present invention. Suppose that switches 122 and 124 operate as VRRP forwarders 712 for fabric switch 120. Switches 122 and 124 can operate as an active and a standby VRRP forwarder, respectively. Similarly, switches 132 and 134 operate as VRRP forwarders 714 for fabric switch 130. Switches 132 and 134 can operate as an active and a standby VRRP forwarder, respectively. Operations of VRRP are described in IETF RFC "Virtual Router Redundancy Protocol (VRRP)," available at http://tools.ietf.org/html/rfc3768.

0086Without VRRP localization, when virtual machine 614 migrates to host machine 620, switch 122 still remains the active forwarder for virtual machine 614. As a result, virtual machine 614 continues to use the IP address of VRRP forwarders 712 as the gateway IP address and forwards packets based on a corresponding MAC address. When switch 138 receives a packet from virtual machine 614, switch 138 forwards the packet to extension switch 134, which, in turn, encapsulates the packet with a tunnel encapsulation and forwards the tunnel-encapsulated packet to fabric IP address 308 of fabric switch 120. Extension switch 122 receives the tunnel-encapsulated packet, decapsulates the tunnel encapsulation, and forwards the inner packet from virtual machine 614.

0087To solve this problem, VRRP forwarders are localized in extended fabric switch 100. As a result, VRRP forwarders 712 operate for the end devices coupled to fabric switch 120 (e.g., virtual machines in host machine 610), and VRRP forwarders 714 operate for the end devices coupled to fabric switch 130 (e.g., virtual machines in host machine 620). Upon completion of the migration of virtual machine 614, extension switch 132 notifies virtual machine 614 that the active forwarder has been updated. Virtual machine 614 then updates the MAC address of the default gateway and continues forwarding via the local VRRP forwarders 714.

<u>Work Load Management</u>

0088<figref idref="f0015">FIG. 8</figref> illustrates an exemplary work load management in an extended fabric switch, in accordance with an embodiment of the present invention. In this example, an Application Resource Broker (ARB) 840 monitors the workload on the virtual machines coupled to fabric switches 120 and 130. If the load on the virtual machines coupled to fabric switch 120 reaches a threshold, ARB 840 can instruct VCenter 632 to create more virtual machines. For example, virtual machine 802 can be created to run on hypervisor 612 in host machine 610 coupled to switch 126, and virtual machines 814, 816, and 818 can be created to run on hypervisor 812 in host machine 810 coupled to switch 128 (denoted with dotted lines). The load on the virtual machines coupled to fabric switch 120 can continue to increase. However, host machines coupled to fabric switch 120 may not have enough computing resources.

0089To solve this problem, ARB 840 is allowed to monitor and mange the workload of extended fabric switch 100 together. As a result, when the load on the virtual machines coupled to fabric switch 120 continues to increase but host machines coupled to fabric switch 120 do not have enough computing resources, virtual machines can be created in host machines coupled to fabric switch 130. For example, virtual machines 824, 826, and 828 can be created to run on hypervisor 822 in host machine 820 coupled to switch 136 (denoted with dashed lines). Since extension switches of extended fabric switch 100 allows communication between the virtual machines, virtual machines 824, 826, and 828 can share configuration (e.g., same VLAN tag) with virtual machines 802, 814, 816, and 818.

<u>Direct Forwarding</u>

0090<figref idref="f0016">FIG. 9</figref> illustrates an exemplary direct forwarding of a data packet in an extended fabric switch, in accordance with an embodiment of the present invention. Direct forwarding allows fabric encapsulated packets to be directly forwarded without being decapsulated at the extension switches. A switch with direct forwarding support can forward in the local fabric switch a fabric-encapsulated packet encapsulated in a remote fabric switch. For example, if switch 132 supports direct forwarding, switch 132 can forward in fabric switch 130 a fabric-encapsulated packet encapsulated in fabric switch 120. In the example in <figref idref="f0016">FIG. 9</figref>, switches in extended fabric switch 100 support direct forwarding.

0091It should be noted that direct forwarding may require a unique switch identifier for a respective switch in an extended fabric switch. To facilitate direct forwarding, a MAC address learned from a neighbor fabric switch is stored in association with a switch identifier of the switch from which the MAC address has been learned. As a result, a respective switch can identify the egress switch identifier of a packet destined to that MAC address. When a switch receives a packet (e.g., an Ethernet frame) via a local edge port, the switch encapsulates the packet with a fabric encapsulation, and assigns the local and the identified switch identifiers as ingress and egress switch identifiers of the encapsulation header, respectively.

0092During operation, end device 142 initiates a communication with end device 148 and sends an Ethernet frame 902 destined to end device 148. Edge switch 126 receives Ethernet frame 902 via an edge port. Suppose that switch 126 has already learned and stored the MAC address of end device 148 in association with switch identifier 354 of switch 138. Switch 126 encapsulates Ethernet frame 902 with a fabric encapsulation (e.g., TRILL or IP encapsulation) to create fabric-encapsulated packet 922 (operation 932). Since switch 126 knows the destination MAC address, switch 126 assigns switch identifier 354 of switch 138 as the egress switch identifier and switch identifier 304 of switch 126 as the ingress switch identifier of the encapsulation header. Switch 126 forwards packet 922 to switch 122.

0093When packet 922 reaches extension switch 122, switch 122 encapsulates packet 922 in a tunnel encapsulation corresponding to tunnel mesh 102 to generate tunnel-encapsulated packet 924 (operation 934). Switch 122 assigns fabric IP address 310 of fabric switch 130 as the destination address and fabric IP address 308 of fabric switch 120 as the source address of the encapsulation header. Switch 122 forwards packet 924 to fabric switch 130 via layer-3 device 112 or 114, as described in conjunction with <figref idref="f0001">FIG. 1</figref>. Since fabric IP address 310 is a floating IP address, switch 132 or 134 can terminate forwarding of packet 924.

0094Suppose that switch 132 receives packet 924, terminates forwarding, and decapsulates the tunnel encapsulation to extract fabric-encapsulated packet 922 (operation 936). Switch 132 forwards packet 922 based on switch identifier 354 to switch 138. Upon receiving packet 922, switch 138 decapsulates the fabric encapsulation to extract Ethernet frame 902 (operation 938). Switch 126 forwards Ethernet frame 902 via the edge port coupling end device 148. End device 148 thus receives Ethernet frame 902 sent from end device 142 based on direct forwarding.

Exemplary Switch

0095<figref idref="f0017">FIG. 10</figref> illustrates an exemplary switch in an extended fabric switch, in accordance with an embodiment of the present invention. In this example, a switch 1000 includes a number of communication ports 1002, a packet processor 1010, a learning module 1030, and a storage device 1050. Packet processor 1010 extracts and processes header information from the received frames.

0096In some embodiments, switch 1000 maintains a membership in a fabric switch, which can further be in an extended fabric switch, as described in conjunction with <figref idref="f0001">FIG. 1</figref>, wherein switch 1000 also includes a fabric switch module 1060. Fabric switch module 1060 maintains a configuration database in storage device 1050 that maintains the configuration state of every switch within the fabric switch. Fabric switch module 1060 maintains the state of the fabric switch, which is used to join other switches. In some embodiments, switch 1000 can be configured to operate in conjunction with a remote switch as an Ethernet switch.

0097Communication ports 1002 can include inter-switch communication channels for communication within the fabric switch. This inter-switch communication channel can be implemented via a regular communication port and based on any open or proprietary format. Communication ports 1002 can also include one or more extension communication ports for communication between neighbor fabric switches. Communication ports 1002 can include one or more TRILL ports capable of receiving frames encapsulated in a TRILL header. Communication ports 1002 can also include one or more IP ports capable of receiving IP packets. An IP port is capable of receiving an IP packet and can be configured with an IP address. Packet processor 1010 can process TRILL-encapsulated frames and/or IP packets.

0098During operation, learning module 1030 identifies from a notification message from a neighbor fabric switch a MAC address learned at the neighbor fabric switch and stores the MAC address in a local MAC table in association with the fabric IP address of the neighbor fabric switch. In some embodiments, switch 1000 can also include a discovery module 1042 which identifies a discovery message comprising the fabric IP address as a source IP address and, in response, includes the IP address in the neighbor list of the extended fabric switch. Switch 1000 can also include switch a high availability module 1044 which operates switch 1000 as an active forwarder of a VRRP forwarder for a virtual machine. This virtual machine has been migrated from a host machine coupled to the neighbor fabric switch to a host machine coupled to the local fabric switch, as described in conjunction with <figref idref="f0014">FIG. 7</figref>.

0099In some embodiments, switch 1000 also includes a forwarding module 1020 which identifies the MAC address as destination MAC address of a packet encapsulated in a fabric encapsulation header. In response, forwarding module 1020 encapsulates the packet (i.e., without he fabric encapsulation header) in a tunnel encapsulation header. Forwarding module 1020 also encapsulates an ARP request in a tunnel encapsulation header. Forwarding module 1020 can further encapsulate a packet destined to a virtual machine, which has been migrated to a neighbor fabric switch, in a tunnel encapsulation header, as described in conjunction with <figref idref="f0013">FIG. 6B</figref>.

0100In some embodiments, forwarding module 1020 identifies a packet encapsulated in a first tunnel encapsulation header corresponding to a tunnel between two hypervisors, as described in conjunction with <figref idref="f0012">FIG. 6A</figref>. Forwarding module 1020 can also identify a packet encapsulated in a fabric encapsulation header, as described in conjunction with <figref idref="f0016">FIG. 9</figref>. In either case, forwarding module 1020 encapsulates the encapsulated packet in a second tunnel encapsulation header corresponding to a tunnel mesh in the extended fabric switch.

0101Note that the above-mentioned modules can be implemented in hardware as well as in software. In one embodiment, these modules can be embodied in computer-executable instructions stored in a memory which is coupled to one or more processors in switch 1000. When executed, these instructions cause the processor(s) to perform the aforementioned functions.

0102In summary, embodiments of the present invention provide a switch and a method for extending fabric switches. In one embodiment, the switch includes a fabric switch module and a learning module. The fabric switch module maintains a membership in a first fabric switch. A fabric switch includes a plurality of switches and operates as a single switch. The first fabric switch is in an extended fabric switch which further comprises a second fabric switch. The learning module identifies from a notification message from the second fabric switch a MAC address learned at the second fabric switch. The learning module stores the MAC address in a local MAC table in association with an IP address of the second fabric switch.

0103Some background for the present disclosure can be found in <patcit id="pcit0002" dnum="US08723911" dnum-type="L"><text>U.S. Patent Application No. 13/087,239</text></patcit>, titled "Virtual Cluster Switching," by inventors Suresh Vobbilisetty and Dilip Chatwani, filed 14 April 2011. The reader of the present disclosure is expected to be familiar with (and have an understanding of) this earlier disclosure.

0104The methods and processes described herein can be embodied as code and/or data, for example as a computer program comprising program instructions in machine-readable format which can be stored in a computer-readable non-transitory storage medium. When a computer system reads and executes the code and/or data stored on the computer-readable non-transitory storage medium, the computer system performs the methods and processes embodied as data structures and code and stored within the medium.

0105The methods and processes described herein can be executed by and/or included in hardware modules or apparatus. These modules or apparatus may include, but are not limited to, an application-specific integrated circuit (ASIC) chip, a field-programmable gate array (FPGA), a dedicated or shared processor that executes a particular software module or a piece of code at a particular time, and/or other programmable-logic devices now known or later developed. When the hardware modules or apparatus are activated, they perform the methods and processes included within them.

0106The foregoing descriptions of embodiments of the present invention have been presented only for purposes of illustration and description. They are not intended to be exhaustive or to limit this disclosure. Accordingly, many modifications and variations will be apparent to practitioners skilled in the art. The scope of the present invention is defined by the appended claims.