A method for transmission of data packets through a network

0001In the following networks are considered, which consist of central nodes and edge nodes. At least some of the central nodes are connected with one or more edge nodes, whereby every edge node is connected with only one central node. The central nodes of the considered network have no storage capabilities. This will be explained later in detail.

0002The edge nodes receive data packets from sources and send data packets to sinks outside the network.

0003In order to transmit data packets through a network, different methods are known.

0004In an edge node of a network, like an Optical Burst Switched (OBS) network, received or incoming data packets, like Internet Protocol (IP) packets, Asynchronous Transfer Mode (ATM) cells or Protocol Data Units (PDUs), are aggregated in the node to a burst with an associated header. The burst contains a number of data packets to a certain destination and is aggregated according to a predefined aggregation strategy. This burst is sent through the network to the destination. Before sending the burst, a path, way or trail has to be set up or reserved in the network, in order that the burst is properly transmitted through the network to its destination. This is done by the associated header. After aggregation of the burst, the header is generated and sent from the edge node to the connected or associated central node. There the header is evaluated and a path is reserved for the succeeding burst. Then the header is sent to a next central node, the header is evaluated, a path is reserved and the header is sent to a next central node until a path to a destination for the succeeding burst is reserved. After sending the header and elapsing of a certain time in the edge node, which is for the path reservation in the central nodes, the burst is sent to the associated central node. The burst is sent from central node to central node through the reserved path to a destination of the burst, which is a certain edge node. There the burst is separated into the data packets, which are sent to further destinations.

0005In networks, like Optical Packet Switching (OPS) networks, the same principles as in OBS networks are applied, with the difference, that every data packet is transformed to a packet with an associated header. The packets are sent like the bursts. The same reservation and transmission principles as applied for the header of the burst are applied for the header of a packet.

0006In principle, in case of an OBS and OPS network no storage capacities for the bursts or packets are provided in the central nodes of the network. They could be provided by fibre delay lines for storing optical packets or bursts. These optical storages are currently very expensive. Limited storage capacities are provided for the header of a burst or packet, as they are much smaller. The header can be converted from optical to electrical, evaluated, stored and modified electrically and converted to optical again.

0007A key characteristic of these networks is the one-pass reservation scheme of network resources respectively network nodes for each individual burst or packet. Bursts are sent without an acknowledgement of successful path set up and burst loss can occur in case of contention.

0008Examples for OBS networks are described in the following documents. In "Application of Tell & Go and Tell & Wait Reservation Strategies in a Optical Burst Switching Network: a Performance Comparision" from A. Detti, M. Listanti, University of Rome, a comparison between two resource reservation techniques when they are used in an All Optical Network for the support of high-speed IP traffic are described.

0009In "Choices, Features and Issues in Optical Burst Switching" from C. Qiao, M. Yoo, University at Buffalo (SUNY), design choices in burst-switching are explored and a new variation that is especially suitable for optical WDM networks is described. The source nodes of an OBS network are adapted to work in the electrical and in the optical domain. According to the Just-Enough-Time protocol a source sends out a control packet, which is followed by a burst after an offset time. The burst is buffered at the source in the electrical domain while the control packet is being processed. If the requested bandwidth is not available, the burst is said to be blocked, and will be dropped or may be buffered and then be transmitted later.

0010In "JumpStart: A Just-in-Time Signalling Architecture for WDM Burst-Switched Networks" from Ilia Baldine, George N. Rouskas, Harry G. Perros and Dan Stevenson, MCNC and North Carolina State University, an architecture for a core DWDM network which utilizes the concept of optical burst switching coupled with a just-in-time signalling scheme is presented.

0011In a ring network, a number of central nodes are connected to a ring, whereby at least a part of the central nodes have a connection to at least one associated edge node, where an edge node is connected with one central node.

0012Any node detecting a failure is adapted to send back a failure message to the source node (page 89, left column, last paragraph). Bursts belonging to the same connection are guaranteed to take the same path, but they are not guaranteed to be successful

0013In a star network, there is only one central node, which has connections to several edge nodes.

0014In a bus network, a number of central nodes are connected in a row, where every central node has two neighbours, except the central nodes at the end of the bus (respectively row). At least a part of the central nodes are connected to at least one edge node, whereby every edge node has a connection to only one central node.

0015It is an object of the invention to present methods for transmission of data packets or bursts through a ring-, star -or bus-network with no packet respectively burst losses.

0016This object is achieved by the features cited in claim 1 or 2.

0017Due to the fact that packets respectively bursts are not sent in case of a reservation or blocking conflict respectively contention in a central node the advantage of the invention is that packet or burst losses are avoided.

0018Further improvements of the invention are identified in subclaims.

0019In an embodiment of the invention an answer is sent from the central node to the edge node with a time or time lag for an occupancy of the central node. Said time or time lag is generated by the central node according to an analyse of its reservation scheme respectively reservation schedule for the data packets or bursts. After the analysis, the central node calculates a time or time lag for the associating edge node, when a new attempt to send a header respectively for the reservation of a path seems to be successful. This time or time lag is inserted in the answer and transmitted to the edge node. The answer is received and evaluated. After expiring of this time or time lag, the edge node sends the previously sent header again to the central node, in order to try to reserve a path for its packet or burst. Meanwhile the packet or burst is stored in the edge node. This has the advantage of an economic attempt for a reservation.

0020In an other embodiment of the invention in case of a reservation conflict the central node schedules a reservation for the burst or packet of the associated header, determines a particular time or a time lag for transmission of said packet or burst, sends an answer where said particular time or time lag is contained to the originating edge node. There said answer is evaluated, the sending of the packet or burst is abandoned, the packet or burst is stored in the edge node and after expiring of said time or time lag the packet or burst is sent to the central node, to use the scheduled reservation. This has the advantage of an economic reservation and loss free transmission of packets or bursts.

0021In a further embodiment of the invention, arriving data packets are added to a stored burst. This has the advantage, that only one burst has to be reserved instead of two successive bursts. A higher multiplexing and transmission gain is achieved, which is economic.

0022Further advantages of the invention are mentioned in the embodiment of the invention.

0023An exemplary embodiment of the invention is explained by means of a drawing.

Shown in the drawing is:

0024Figure 1 a ring network.

0025Figure 1 shows ten central nodes respectively ring nodes node_0, node_1, ..., node_9, whereby every central node is connected with two other central nodes to form a ring network. These central nodes can be switches, network elements, routers or optical switches for an OBS or OPS network. Five edge nodes Edge_0, Edge 1, ..., Edge 4 are connected to a central node. Edge node Edge_0 is connected with central node node_0, edge node Edge_1 is connected with central node node_2, edge node Edge_2 is connected with central node node_4, edge node Edge_3 is connected with central node node_6 and edge node Edge_4 is connected with central node node_8.

0026The edge nodes Edge_0, Edge 1, ..., Edge 4 aggregate data packets from not shown sources to bursts, like optical bursts, or transform them to optical packets as explained before. These burst/packets are sent from the edge node to the associated central node and further through the central network to a destination edge node. The edge nodes Edge_0, Edge 1, ..., Edge 4 divide bursts received by the OBS or OPS network respectively central nodes into data packets and send them to not shown sinks or destinations.

0027In such a ring network, the only nodes where blocking might take place are those which are connected to an edge node. These are the central nodes node_0, node_2, node_4, node_6 and node_8. The rest of the central nodes, e.g. node_1, node_3, node_5, node_7 and node_9, merely forward the packets or bursts and therefore are blocking free.

0028After sending a header of a burst or packet from an edge node to an associated central node / ring node two cases can happen.

0029First, the resources for transmitting the burst/packet are available in the central node and a path is reserved according to the already described behaviour. After an offset time the burst/packet in the edge node is sent and the burst/packet is transmitted through the reserved paths to its destination.

0030Second, the resources for reservation of a path and transmitting the burst/packet are not available. To avoid a collision, an answer is sent back to the originating edge node causing the burst to wait, until further transmission attempt(s) take place. After sending of the header from the edge node to the associated central node, the header is evaluated in the central node and the requested reservation is compared to previous reservations of other headers. In case of a reservation conflict an answer is sent back to the originating edge node. Said answer is evaluated in the edge node and the sending of the packet or burst is abandoned and the packet or burst is stored in the edge node in the electrical domain.

0031The answer comprises the time or time lag of occupancy or a time or time lag for a future reservation or any other kind of message to transfer an information from the edge node to the central node.

0032According to the information sent to the edge node and an implemented mechanism, this node can have a different behaviour.

0033First, in case of a reservation conflict the central node determines a time or a time lag for an occupancy of the central node and sends an answer, where said time or time lag is contained, to the originating edge node. Said answer is evaluated in the edge node, the sending of the burst or packet is abandoned, the burst or packet is stored in the edge node and after expiring of said time or time lag a header is sent again to the central node, to try a new reservation for a transmission of said burst or packet.

0034Second, that in case of a reservation conflict the central node schedules a reservation for the burst or packet of the associated header, determines a particular time or a time lag for transmission of said burst or packet and sends an answer, where said particular time or time lag is contained, to the originating edge node. There said answer is evaluated, the sending of the burst or packet is abandoned, the burst or packet is stored in the edge node and after expiring of said time or time lag the burst is sent to the central node, to use the scheduled reservation.

0035It is independent, which burst reservation mechanisms, like just-in-time (JIT), Horizon reservation mechanism (Horizon) or just-enough time (JET), is used. If a header of a burst/packet from an edge node arrives at the central node, the node knows according to its reservation schedule until when it will be busy. This is because of old reservations by other headers received from neighbour central nodes. The central node can therefore deduce if the optical burst/packet associated to the header will be blocked or not.

0036During the storing of the burst in the edge node arriving data packets can be aggregated to the stored burst. If this is done, a second reservation for a burst, which contains in the meantime arriving data packets, is avoided. Therefore only one burst, which is larger than before and has a higher multiplexing gain, is sent. Because one larger burst is sent instead of two bursts with corresponding headers, less resources are used.

0037Once a burst/packet is inserted into the ring, star or bus network, it will continue its way to the destination edge node without any kind of blocking. If the network load increases, optical bursts / packets that can not be transported by the network will be stored in the edge nodes. If the network load reaches a critical level, the buffers in the edge nodes will eventually overload and lose information. However, even under these heavy load conditions the network (ring, star, bus) will be loss-free.

0038In the edge node the mentioned burst can be transformed and sent as an optical burst to a central node of an Optical Burst Switched network. Also, in the edge node the mentioned packet can be transformed and sent as an optical packet to a central node of an Optical Packed Switched network.

The main principles of the invention are:

0039The nodes, at which blocking might take place send back an answer to the correspondent edge node if they foresee according to their burst reservation schedule that its associated burst or packet will be blocked.

0040If an edge node receives said answer, the sending of the burst/packet is abandoned and the burst/packet is stored. In case of an OBS network this might happen by storing an electrical version of the optical bursts/packets which have to be sent to the ring. Often such edge nodes are an interface between the electrical and optical domain.

0041If an edge node sends a header and receives an answer back, it knows automatically that the associated burst/packet will not get through the next central node due to blocking. It is also possible that the answer of a burst/packet that is about to be blocked carries some useful information from the central node on its way back to the edge node. This information could be for instance, the exact moment when the central node is reserved for the optical burst/packet. Knowing this, the edge node can immediately send the optical burst/packet right when the time/moment is achieved.

0042The same principles described above by an example for a ring network can be analogue applied to a star (one central node with many edge nodes) and a bus network ("split or broken ring") without modifications.

0043The inventive methods extend the principles of burst or packet switched networks, like OBS and OPS networks, with central node without burst or packet storage capabilities from the unacknowledged sending of a burst to a "backpressure" communication for the first hop from the edge node to the central node in case of conflict in the central node.

0044Further advantages of the inventive methods are: <ul id="ul0001" list-style="bullet" compact="compact"><li>It reduces the blocking probability in the network to zero.</li><li>The solution is especially suited for Optical Burst and Packet Switched networks.</li><li>The solution is especially suited for ring, star and bus networks.</li><li>It is simple and easy. It does not demand much processing time neither at the central nodes nor at the edge nodes.</li><li>It is not a bandwidth consuming, since it does not require to increase the size of the optical headers.</li><li>It is cheap since it requires no additional hardware. In addition for OBS / OPS networks storing the information electrically in the edge nodes is a low priced solution compared to optical storing with Fibre Delay Lines (FDL).</li><li>The method can be implemented in software.</li></ul>