Network attached virtual data storage subsystem

[0001] This is a continuation-in-part application of prior application Ser. No. 09/343,512 filed on Jun. 30, 1999, which is a continuation of prior application Ser. No. 08/815,739 filed on Mar. 12, 1997.

[0002] 1. Field of the Invention

[0003] This invention relates to data storage systems and, in particular, to a plurality of tape devices and automated tape libraries where the tape devices are connected to a plurality of data processors via a high speed network and which along with a Network Storage Manager system controller collectively implement a virtual, distributed data storage subsystem. This readily scalable data storage subsystem creates multiple virtual storage devices on demand, which are available to any and all of the system's data processors.

[0004] Problem

[0005] It is a problem in the field of data storage subsystems to provide their attached data processors with economical data storage services at adequate performance levels. Although storage device technology can support the development and implementation of specialized devices with uniquely high performance levels, such devices are generally not economically feasible. Commercial, off-the-shelf storage devices are typically the most economical building blocks for data storage subsystems but these devices can impose performance limitations; applications requiring performance levels beyond the storage device's native characteristics require the use of architectural approaches.

[0006] One other aspect of this data storage problem is that the allocation of customer data to a single type of data storage device represents a limitation when faced with widely varying data storage needs. This limitation can be partly obviated by sharing input/output (I/O) activity and data bandwidth across a plurality of data storage devices of a data storage subsystem. However, a fixed configuration of data storage devices also limits the scalability in performance and provides no facility for applications to request changes in data storage performance. An architecture where the data storage devices are located behind a server(s) further limits the delivered performance since the data storage bandwidth is limited by the server itself. Therefore, architecting a data storage subsystem that can efficiently serve the needs of the applications extant on the data processors is a daunting problem. There are numerous factors that affect performance; this problem is particularly pertinent to tape devices, since the tape devices must serve a wide range of performance requirements and it would be desirable for a single tape device type to serve this wide range of needs.

[0007] In FIG. 1, a traditional “direct attach” tape configuration is shown. Here tape devices TD<b>1</b>, TD<b>2</b>, TD<b>3</b> are each directly connected to single data processors DP<b>1</b>, DP<b>2</b>, DP<b>3</b> on a control network CN in a dedicated tape device configuration. The data processor has exclusive use of the tape device and typically communicates with the tape device via a SCSI or Fibre Channel interface. However, the use of dedicated tape devices is an expensive proposition where there are a plurality of data processors to be served, especially if the data access loads generated by the plurality of data processors are erratic. In this data storage subsystem architecture, the utilization of the tape devices and the efficiency of the data storage function are less than optimal, since each data processor is limited to its dedicated tape device with its physical constraints.

[0008]FIG. 1 also includes an automated tape cartridge library system (ACS) and an automated cartridge controller (ACSC) which provides enhanced response time for the tape devices by mounting/dismounting their tape cartridges. However, the tape devices may have a SCSI interface in the data path that introduces a number of physical limitations to the operation of the system. The first limitation is that only a small number of tape devices can be attached to a SCSI bus compared to other bus architectures. The second limitation is the limited bandwidth of the SCSI bus that is shared by these tape devices. The length of the SCSI cable also represents a limitation, since it is typically restricted to 25 meters.

[0009] An alternative data storage subsystem architecture is shown in FIG. 2. Here, a traditional “client/server” network attached tape configuration includes a plurality of tape devices TD<b>1</b>, TD<b>2</b>, TD<b>3</b> each attached to the data communication network N via dedicated servers TS<b>1</b>, TS<b>2</b>, TS<b>3</b>; data processors DP<b>1</b>, DP<b>2</b>, DP<b>3</b> connect to the network directly. In this architecture, the data processors all have access to all of the tape devices via their servers. The data processors run tape server software to manage the access protocol for the plurality of tape servers. Though this configuration allows tape devices to be allocated dynamically, it does not scale very well with the demand on the data transfer rate because the bandwidth is limited by the server.

[0010] A variation of this storage architecture is the use of a plurality of tape devices configured into an array. As shown in FIG. 3, the tape devices are configured in a Redundant Array of Independent Tapes (RAIT) in a manner analogous to the Redundant Array of Independent Disks (RAID) which is a well known architecture in the disk device technology. Using a RAIT controller (RC), the tape array is typically located behind a server(s) RS<b>1</b>, RS<b>2</b>, which is directly connected to the network N The bandwidth for data transfers between the data processors and the tape array is not scalable and is also limited by the characteristics of the server(s). The tape array itself is also not scalable or easily reconfigured due to the limitations of the server(s).

[0011] These various tape data storage subsystem architectures are all limited in their data exchange I/O activity and bandwidth to that of a single tape device or a fixed RAIT device. The alternative is to use the integration of a plurality of data storage media types (e.g. solid state memory, disk, tape, etc.) and data storage subsystem architectures into a single data storage subsystem. However, this results in increased storage system complexity and costs.

[0012] Solution

[0013] The above-described problems are solved and a technical advance achieved in the field by the network attached virtual tape data storage subsystem of the present invention. This invention relates to tape data storage systems and, in particular, to a plurality of tape devices and automated tape libraries where the tape devices are connected to a plurality of data processors via a high bandwidth network and which, along with a Network Storage Manager (NSM) controller, collectively implement a virtual, distributed tape data storage subsystem. The virtual, distributed tape data storage system creates multiple virtual tape devices on demand by allocating, configuring and managing the system's individual physical tape devices. The I/O rate and bandwidth of the individual virtual tape devices are changeable on demand; the aggregate performance of the virtual, distributed tape data storage subsystem is readily scalable by changing the number of physical tape devices and library devices. By pooling the tape devices together, interconnecting them with the data processors via a network, incorporating the Network Storage Manager controller and creating virtual tape devices, the problems of prior art tape device data storage subsystem architectures are overcome. This new architecture can create simultaneous, multiple virtual devices, which are available to any and all of the data processors—subject only to the limitations of the aggregate capacities and bandwidth of the tape devices, library devices and the network. This architecture enables the storage requirements of the data processors to be met by both individual tape devices or multiple devices in any tape array configurations—including, but not limited to: RAIT <b>0</b>, <b>1</b>, <b>3</b>, <b>4</b>, <b>5</b>. The tape array and data transmission bandwidth can be dynamically reconfigured on demand since the network directly interconnects the tape devices to the data processors.

[0014] The virtual tape storage subsystem is controlled by a Network Storage Manager (NSM) which includes a plurality of software elements including: Resource Allocation (RA) software, Resource Configuration (RC) software, Resource Management (RM) software, and Security Management software (SM). The NSM communicates with the Automatic Cartridge System Library Software (ACSLS) via the RA software. The ACSLS communicates with the ACS via network CN. The RA software has the responsibility to keep track of the resource usage such as which data processor presently owns which tape device(s), which tape devices are free to be allocated to the requesting data processors and the like. The RC software configures the data storage resources for the network attached data processors—responding to manual operator commands, pre-programmed algorithms, rules, application program initiated requests and the like. The RC software assigns the maximum number of tape devices that a data processor can use and designates the configuration of these tape devices. Furthermore, it automatically configures the tape devices allocated to a data processor as well as the connection between the data processor and the tape device(s). The RM software queues the request for the resource allocation and notifies the data processor when the requested resource is ready, or it can schedule the availability of the resources. The ACSLS has the responsibility to keep track of the tape cartridges that are stored in the ACS. It also keeps track of the status of the tape devices, and provides the application programming interface (API) to other application software. The API will allow the application software to load tape or to mount a volume to a tape drive, to unload tape or to unload a volume from a tape drive, to read the status of tape devices, etc.

[0015] The use of a networked storage manager enables the dual function of permitting tape devices to be managed as a pool and yet be attached directly to the network as individual resources. In addition, the networked storage manager provides the mechanism for the central control and management of the resources attached to the network such as to control tape device allocation and configuration, as well as other functions, such as tape cartridge movement and data migration. The rules, which could be implemented, address response time constraints, data file transfer size, data file transfer rates, data file size bounds and the like.

BRIEF DESCRIPTION OF THE DRAWINGS

[0016]FIG. 1 illustrates in block diagram form the overall architecture of a traditional direct attached tape configuration;

[0017]FIG. 2 illustrates in block diagram form the overall architecture of a traditional client/server network attached tape configuration;

[0018]FIG. 3 illustrates in block diagram form the overall architecture of a traditional network attached RAIT configuration; FIG. 4 illustrates in block diagram form the overall architecture of an embodiment of the network attached virtual tape storage subsystem of the present invention;

[0019]FIG. 5 includes a detail of the software architecture for the ACSC; and

[0020]FIGS. 6, 7 and <b>8</b> illustrate in block diagram form the overall architecture of the dynamic mapping and data striping functions of the network attached virtual tape storage subsystem of the present invention.

DETAILED DESCRIPTION

[0021] Existing Tape Data Storage System Architectures

[0022]FIG. 1 illustrates in block diagram form the overall architecture of a traditional direct attached tape device configuration, wherein a plurality of data processors DP<b>1</b>-DP<b>3</b> are each directly connected to at least one tape device TD<b>1</b>-TD<b>3</b>, which tape devices TD<b>1</b>-TD<b>3</b> for the plurality of data processors DP<b>1</b>-DP<b>3</b> are served by an automated tape cartridge library system ACS and its associated automated tape cartridge library system controller ACSC. in practice, the ACS consists of one or more robotic library units. The data processors DP<b>1</b>-DP<b>3</b> communicate with the automated tape cartridge library system controller ACSC via network CN. The automated tape cartridge library system controller ACSC receives data file access requests from the data processors DP<b>1</b>-DP<b>3</b> and allocates a designated tape cartridge located in the automated tape cartridge library system ACS to serve the request. The automated tape cartridge library system controller ACSC transmits commands to the automated tape cartridge library system ACS to retrieve the selected tape cartridge from its cartridge storage location in the automated tape cartridge library system ACS and mount the retrieved tape cartridge on the tape device TD<b>1</b>-TD<b>3</b> directly connected to the requesting data processor DP<b>1</b>-DP<b>3</b>. The automated tape cartridge library system controller ACSC transmits commands to the requesting data processors DP<b>1</b>-DP<b>3</b> when the selected tape cartridge is mounted in the identified tape device TD<b>1</b>-TD<b>3</b> to indicate that the requested data is presently available on the identified tape device TD<b>1</b>-TD<b>3</b>.

[0023]FIG. 2 illustrates in block diagram form the overall architecture of a traditional client/server network attached tape configuration that is similar in operation to the system of FIG. 1. The differences are that the data processors DP<b>1</b>-DP<b>3</b> (the clients) are no longer directly connected to the tape devices TD<b>1</b>-TD<b>3</b>, but share the plurality of tape devices TD<b>1</b>-TD<b>3</b> among the data processors DP<b>1</b>-DP<b>3</b>. The communications among the data processors DP<b>1</b>-DP<b>3</b>, tape devices TD<b>1</b>-TD<b>3</b> and automated tape cartridge library system controller ACSC are effected via a common network N which transmits both data and control information there between. Tape devices, such as those with SCSI interfaces, typically lack the ability to communicate directly with the network. To accommodate the network connection, tape devices TD<b>1</b>-TD<b>3</b> are all located behind corresponding server processors TS<b>1</b>-TS<b>3</b> connecting with them using a SCSI interface. This is a fixed, static configuration, difficult to scale to changing data storage needs.

[0024]FIG. 3 illustrates in block diagram form the overall architecture of a traditional network attached RAIT array configuration where the individual tape devices of FIG. 2 are replaced by an array of tape devices TD<b>1</b>-TD<b>4</b>. The tape devices TD<b>1</b>-TD<b>4</b> are configured in a redundant array of data storage devices in a manner analogous to the Redundant Array of Independent Disks (RAID) which is a well known architecture in the disk device technology. The operation of a RAID system, and by extension, the corresponding operation of the RAIT system is not described in detail herein, since such information is readily available in the literature. The RAIT system includes a RAIT controller RC that is connected to the network N via a plurality of servers RS<b>1</b>-RS<b>2</b>. The traditional RAIT is located behind a server, using a SCSI or Fibre Channel interface, and is a further example of a fixed, static configuration, difficult to scale to changing data storage needs.

[0025] System Architecture of Distributed Virtual Tape Storage Subsystem

[0026] The architecture of the preferred embodiment as shown in FIG. 4 uses a shared pool of tape device resources TD<b>1</b>-TD<b>5</b> to support multiple virtual storage devices, which are available to any and all of the data processors DP<b>1</b>-DP<b>4</b>. The number and performance levels of the virtual devices are only limited by the aggregate capacity and bandwidth of the tape device pool and the interconnecting network FN. This architecture also enables any tape array configuration (including, but not limited to RAIT <b>0</b>, <b>1</b>,<b>3</b>, <b>4</b>, <b>5</b>) to be realized using the pooled tape devices TD<b>1</b>-TD<b>5</b>. The tape array(s) can be reconfigured dynamically since network FN interconnects the tape devices TD<b>1</b>-TD<b>5</b> to the data processors DP<b>1</b>-DP<b>4</b> under control of a network storage manager NSM whose functions are contained in the ACSC controller. The bandwidth of this networked distributed virtual tape device data storage subsystem is scalable and can be changed on demand. In addition, network security can be implemented.

[0027] Included in FIG. 4 are connection blocks CB<b>1</b>-CB<b>9</b>. These blocks are where the translation from the physical devices to the desired virtual device occurs. For example, if data processor DP<b>1</b> requires a storage device of a data rate that the physical devices TD<b>1</b>-TD<b>4</b> can not achieve individually, DP<b>1</b> would request, over the network CN, that the NSM (located in the ACSC controller) construct a virtual device with the required date rate. The NSM would communicate with CB<b>1</b> via the CB<b>0</b> link to instruct it to stripe the data requested coming from DP<b>1</b> across devices TD<b>1</b>+TD<b>2</b>+ . . . +TD<b>5</b> as needed to achieve the required data rate. DP<b>1</b> communicates to its side of CB<b>1</b> as though it were one storage device (the virtual device). Within CB<b>1</b>, the data from DP<b>1</b> is striped and sent to the actual physical storage devices over the network FN. In this example the translation from physical to virtual device occurred within CB<b>1</b>; the connection blocks CB<b>5</b>-CB<b>9</b> served only a pass-through function. The connection block CB<b>0</b> links the ACSC interface to the FN network if needed. The connection block CB<b>1</b>-CB<b>4</b> functions can be housed in separate physical boxes or located inside of DP<b>1</b>-DP<b>4</b> (on a network interface-like card) or integrated into the network FN's hardware/software, such as within a networks switch fabric. In a similar manner, connection blocks CB<b>5</b>-CB<b>9</b> could be housed in separate physical boxes or integrated into the storage device's TD<b>1</b>-TD<b>5</b> hardware/software. Likewise, connection block CB<b>0</b> could be housed in a separate physical box or integrated into the hardware/software of the ACSC. Connection blocks CB<b>0</b>-CB<b>9</b> can include the functionality to bridge between dissimilar interfaces such as SCSI and fibre channel thereby permitting SCSI attached storage devices and data processors to be used on a fiber channel network. In an alternate implementation the translation from physical to virtual device can occur in connection blocks CB<b>5</b>-CB<b>9</b> and in this case CB<b>1</b>-CB<b>4</b> serve a pass through function. In this implementation the operation is similar to the previous case except that the data processor DP<b>1</b> would communicate with an assigned connection block for example CB<b>5</b> via CB<b>1</b> via network FN. CB<b>5</b> would now serve the same function as CB<b>1</b> did in the previous case.

[0028] In operation, the data processors DP<b>1</b>-DP<b>4</b> are supported by one or more virtual tape devices. The virtualization process is achieved by the functions in the Connection Blocks CB<b>1</b>-CB<b>9</b>. These functions are implemented in software and/or hardware. Each virtual tape device can be configured dynamically to have different characteristics at the request of the data processors DP<b>1</b>-DP<b>4</b>, as will be further described below. The data processors DP<b>1</b>-DP<b>4</b> do not need to know how the virtual tape device is realized, since their interface to CB<b>1</b>-CB<b>4</b> is consistent irrespective of the virtual tape device being realized.

[0029] Referring to FIG. 5, when an application running under one of the data processors (e.g. DP<b>1</b>) needs a tape device, it makes a request to the RM software to mount a tape volume with the characteristics defined in the request. The request includes data transfer rate, data access time, data compression, and the like. The RM software will communicate with the ACSLS via the RA software to determine if the tape volume is already created or if the tape volume is new. If the tape volume is in the volume data base that ACSLS maintains, the RM software will retrieve the characteristics of this tape volume and determine how many physical tape devices are needed, and what configuration is required to accomplish the requested characteristics. If the tape volume is not in the volume database, the RM software will determine how many physical tape devices are needed based upon the request information. Then, the RM software makes a request to the RA software to reserve that required number of physical tape devices. The RA software communicates with the ACSLS to reserve that number of physical devices and to load the requested tape volume. The ACSLS communicates with one or more (if applicable) ACSs to load the desired tapes to the corresponding tape devices. The RM software instructs the RC software to configure all Connection Blocks. Continuing the example, the RC software instructs the Connection Blocks CB<b>5</b>, CB<b>6</b>, and CB<b>7</b> that they are reserved to transfer data to and from Connection Block CB<b>1</b>. The RC software also instructs the Connection Block CB<b>1</b> that the data that it receives from data processor DP<b>1</b> will be striped to three different physical tape devices TD<b>1</b>, TD<b>2</b>, TD<b>3</b> via the Connection Blocks CB<b>5</b>, CB<b>6</b>, and CB<b>7</b>. The RM software then informs the data processor DP<b>1</b> that the virtual tape device is ready for data transfer. The data processor DP<b>1</b> starts transferring data to the Connection Block CB<b>1</b>, which will buffer the data and transfer it to the physical tape devices TD<b>1</b>, TD<b>2</b>, TD<b>3</b> via CB<b>5</b>, CB<b>6</b>, CB<b>7</b>. When the data processor DP<b>1</b> is done with the data transfer and it wishes to release the virtual tape device, it makes a request to the RM. The RM software instructs the Connection Block CB<b>1</b> to stop transferring data to/from the data processor DP<b>1</b> and then communicates with the RA software to release the physical tape devices. The RA software instructs the ACSLS to unload the volume and to unload the tape cartridges from the tape devices. Finally, the RM software informs the data processor DP<b>1</b> that the virtual device is released.

[0030] The use of the NSM enables the storage devices to be managed as a pool and yet be attached to the network FN as individual resources. The networked storage manager NSM provides the mechanism to control storage device allocation and configuration as well as other functions, such as cartridge movement and data migration. The rules can include response time constraints, data file transfer size, data file transfer rates, data file size bounds and the like. The networked storage manager NSM manages the allocation, configuration and security. Security can be implemented by establishing authentication between the data processors and the NSM to control file access or volume access. In addition, data encryption/de-encryption can be implemented within the control blocks CB<b>1</b>-CB<b>4</b> and activated by a key held by the data processors DP<b>1</b>-DP<b>4</b>.

[0031] Fiber Channel Network

[0032] The SCSI interface is a commonly used bus topology. All devices that are connected to the SCSI bus share the fixed bus bandwidth, with only a single device being capable of accessing the bus at a time—SCSI bus bandwidth is not scalable. As more devices are attached to the SCSI bus, the contention for the bus increases and data transfer performance degrades. In contrast to the SCSI bus, in the fiber channel switched point to point fabric architecture, each branch is capable of the full fiber channel bandwidth; this is a bandwidth scalable system—adding branches increases the aggregate bandwidth of the fiber channel fabric. Although the SCSI interface is not designed for switched point-to-point operation, it can be converted to this use by the addition of a SCSI-to-fiber bridge. Referring to FIG. 4, in order to incorporate the SCSI interface and bus into this distributed virtual tape storage subsystem architecture, each SCSI attached device (DP<b>1</b>-DP<b>4</b>, TD<b>1</b>-TD<b>5</b>, ACSC) is equipped with a SCSI to fiber channel bridge function within connection blocks CB<b>0</b>-CB<b>9</b>.

[0033] The fiber channel network FN is controlled by the NSM software that is running in the automated cartridge library system controller ACSC. The ACSC controller is connected to the fiber channel network FN through CB<b>0</b>. The automated cartridge library system controller ACSC and all of the data processors DP<b>1</b>-DP<b>4</b> are interconnected via an Ethernet control network CN. Since only control information is passed along this network, it does not limit data transfer performance of the overall system.

[0034] In an alternative embodiment of this invention, all control information from network CN is carried by network FN thus eliminating the need for the CN network. In another alternative embodiment, the ACS would attach to the network FN either directly (if fiber channel compatible) or through an appropriate bridge.

[0035] Referring to FIG. 5, in operation, a data processor (e.g. DP<b>1</b>) initiates a request to access a selected data volume that is managed by the ACSLS. DP<b>1</b> then transmits this request, via control network CN, to the RM software that resides in the automated cartridge library system controller ACSC. The RM software receives the request and ascertains the physical location of the physical media that contains the selected data volume. The RM software also identifies an available tape device (e.g. TD<b>4</b>) via the RA software and the ACSLS software. The RM software instructs the ACSLS software via the RA software to transmit control signals to the automated tape cartridge library system ACS to initiate the mounting of the physical media, containing the selected data volume, onto tape device TD<b>4</b>. Once the tape mount is successful, the ACSLS software notifies the successful mount status to the RM software. The RM software then instructs the RC software to transmit a “RESERVED” command via CB<b>0</b>, with the network address form &lt;HOST SCSI ID&gt;&lt;HOST BRIDGE ID&gt; to TD<b>4</b>'s SCSI-to-fiber channel bridge function located in CB<b>8</b> to reserve TD<b>4</b> for the data processor DP<b>1</b>. The RC software also transmits a “TARGET BRIDGE ID” command to the SCSI-to-fiber channel bridge function located in CB<b>1</b> that serves the requesting data processor DP<b>1</b> to thereby identify TD<b>4</b>'s SCSI-to-fiber channel bridge. Once the RM software reserves tape device TD<b>4</b> for the requesting data processor DP<b>1</b>, it returns a status message to the data processor DP<b>1</b> indicating a successful status. The requesting data processor DP<b>1</b> can now transmit read/write and other SCSI commands to the identified tape device TD<b>4</b> via the fiber channel network FN. When the requesting data processor DP<b>1</b> completes the desired operations, it transmits a “DISMOUNT” command via the control network CN to the RM software. Upon receipt of the “DISMOUNT” command, the RM software instructs the ACSLS via the RA software to dismount the tape. The RM software also instructs the RC software to transmit a “RELEASE” command via CB<b>0</b> to the CB<b>8</b> SCSI-to-fiber channel bridge function. When the tape is unloaded, the RA software de-allocates the identified tape device TD<b>4</b> in its resource table and returns a successful status to the RM software. The RM software then returns a successful status to the DP<b>1</b>. The identified tape device TD<b>4</b> is then available to be used by any data processor.

[0036] Tape Array Features

[0037]FIGS. 6 and 7 and <b>8</b> illustrate in block diagram form the overall architecture of the dynamic mapping and data striping functions, respectively, of the network attached virtual tape storage subsystem. In particular, FIG. 6 illustrates the concept of supporting the storage requirements of the data processors DP<b>1</b>, DP<b>2</b> with a plurality of virtual storage devices. In this example, the virtual storage devices include those made up of a “single independent drive” SID, a RAIT <b>3</b> group with three tape devices, a RAIT <b>1</b> group with two tape devices, and a RAIT <b>5</b> group with three tape devices. A plurality of tape devices TD<b>1</b>-TD<b>5</b> are in a pool <b>100</b> that provides resources for the above virtual device configurations. The creation of virtual devices is limited to the tape device resources of the pool; not all of the virtual devices of FIG. 6 can simultaneously exist. The configuration of the tape devices TD<b>1</b>-TD<b>5</b> is controlled by the networked storage manager NSM on a dynamic basis responding to the storage needs of the data processors DP<b>1</b>, DP<b>2</b>. These storage needs may be communicated to the NSM by manual operator commands, preprogrammed algorithms, rules, application program initiated requests and the like.

[0038] Tape Array Configuration

[0039] When a storage device request, issued by a data processor within the system, can be met by the capabilities of a single tape device in the storage device pool, that device (if available) will be temporarily assigned to the requesting data processor. However, if the data processor's storage request exceeds the performance capabilities of a single device in the storage device pool, the system will attempt to allocate multiple storage devices and configure them as a RAIT group. The specific configuration will be dependent upon the storage performance requested.

[0040]FIG. 7 illustrates the concept of striping the data across a plurality of the tape devices TD<b>1</b>-TD<b>5</b>. In this example, a host data processor DP<b>1</b> connects to the connection block CB<b>1</b> via the SCSI ID <b>1</b> interface. Host DP<b>1</b> makes a request to the NSM (not shown in FIG. 7) that it desires to transfer data at some speed “X Mbytes/sec” and that it desires the data to be stored with parity. The NSM translates the desired bandwidth that the host DP<b>1</b> requested into the number of tape devices that support the requested bandwidth. In this example, the number of drives is three: two drives for data and one drive for parity. The NSM allocates drives TD<b>2</b>, TD<b>3</b> and TD<b>4</b> for host DP<b>1</b>. The NSM configures the connection block CB<b>1</b> to stripe the data that comes from SCSI ID <b>1</b> for the write operation and to merge the striped data for the read operation. The NSM also instructs the CB<b>1</b> to send the striped data to the appropriate destination fibre channel port <b>5</b>, <b>8</b> and <b>2</b>. For the write operation, the data received at SCSI ID <b>1</b> is striped into blocks B<b>1</b>, B<b>2</b>, . . . Bn, Bn+<b>1</b>. The connection block CB<b>1</b> sends block B<b>1</b>, B<b>3</b>, . . . , Bn to TD<b>2</b> and TD<b>4</b> via CB<b>6</b>, CB<b>8</b> using the multicast capability of the network FN. It then sends the block B<b>2</b>, B<b>4</b>, . . . , Bn+<b>1</b> to TD<b>3</b> and TD<b>4</b> via CB<b>7</b>, CB<b>8</b> using the same method. Blocks B<b>1</b> and B<b>2</b>, B<b>3</b> and B<b>4</b>, . . . Bn and Bn+<b>1</b> received at the connection block CB<b>8</b> are exclusive OR'ed together to generate the parity data which is sent to TD<b>4</b>. Referring to FIG. 8, for the read operation, data blocks B<b>1</b>, B<b>3</b>, . . . , Bn, B<b>2</b>, B<b>4</b>, . . . , Bn+I and the parity blocks are delivered to the connection block CB<b>1</b>. Data blocks B<b>1</b> and B<b>2</b>, B<b>3</b> and B<b>4</b>, . . . , Bn and Bn+<b>1</b> are merged together in the connection block CB<b>1</b> and the resulting data stream is transferred to the SCSI ID <b>1</b> for delivery to DP<b>1</b>. If one of the data blocks is bad, the connection block CB<b>1</b> will perform data recovery by performing an exclusive OR operation on the corresponding parity block and the good data block to recover the bad data block. The recovered data block and the good data block are merged together and the resulting data stream is transferred to the SCSI ID <b>1</b>.

[0041] Although the preferred system embodiment presented here utilizes tape devices for storage, the same system architecture could be applied to other removable storage device types e.g. optical disk, magnetic disk, optical tape, memory cards, and the like. Furthermore, those skilled in the art would be able to utilize this invention for fixed media storage devices as well e.g. magnetic disk. With fixed media storage devices, the system can be simplified by the omission of robotic libraries and the software used to allocate, configure and manage the plurality of media associated with each storage device.