Available in both "Just a Bunch Of Disks" (JBOD) or "Redundant Array of Independant Disks" (RAID) configurations, the TP9100 offered 1Gb/s (and later 2Gb/s) Fiber Channel storage for the SGI Origin 3000 and SGI Origin 300 series servers.
When the TP9100 Storage System was initially released in 1999, Silicon Graphics offered both RAID (single- and dual-loop) and JBOD (single- and dual-loop) enclosures, each containing 12 Fiber Channel disks and offering 1Gb/s connectivity to a host or SAN.
Both enclosures were offered in both deskside and 4U rackmount configurations. They both provided redundant power supplies and Enclosure System Interface/Operator (ESI/ops) panels allowing full redundancy for all components in the enclosure. These enclosures where essentially a rebranded version of the Xyratex SS-1201-FCAL shelf and, in addition to being part of the TP9100, were sold under the moniker "D-brick" for SGI Origin 3000 systems.
In 2002, Silicon Graphics released a new version of the TP9100 Storage System. Noteably, this new release increased the throughput of the array to 2Gb/s, the number of disks per enclosure to 16 and the number of disks per RAID array to 64. These new enclosures were also rebranded Xyratex units, part number RS-1600-2G, and were aliased as "D-brick2" for SGI Origin 3000 systems.
The RAID enclosures utilized either one or two RAID controllers and could independantly control up to 32 drives (later this was increased to 64), thus allowing up to two expansion enclosures to be cabled to it.
When two RAID controllers are installed in a single enclosure, the system is referred to as a dual-active system and offers uninterrupted access to the RAID array(s) in the case of a single controller failing. Conversely, an enclosure with only a single RAID controller is referred to as a simplex system and does not offer any redundancy at the controller level.
The individual RAID controllers are Mylex intelligent caching controllers supporting RAID levels 0, 1, 3, 5, 1+0 and JBOD. The controller is also capable of RAID levels 10, 30 and 50 in spanned disk arrays.
The cache on each RAID controller consists of 128MB of battery-backed memory. In dual-active configurations, the two controllers maintain coherency in their caches for failure resiliency. When Silicon Graphics shipped a RAID controller, the battery was disconnected, and users are encouraged to ensure that the battery is connected when installing a new RAID controller in an enclosure.
The array is managed by a product called Xyratex Distributed Array manager (XDAM). SGI sells a branded version by the name TPM. All of these configurators were written by David Lethe of SANtools, Inc. As neither SGI (bankrupt 2009) or Xyratex sell or support this product, SANtools is free to supply a compatible configurator  for users of SGI and non-SGI branded subsystems. In addition, data recovery services , and critical firmware updates are also available from SANtools.
A JBOD enclosure was offered either as an expansion module to the RAID enclosure or a standalone system. These enclosures differed from the RAID enclosures only in the lack of RAID controllers and, as such, shared all other components with the RAID enclosures.
The disks utilized in the TP9100 system were standard, hot-swappable Fiber Channel disks and were mounted in sleds for easy addition and removal from the system.
When used in a JBOD configuration, the disk drives were maintained by the operating system and were no different than any standard SCSI disk. However, when used in a RAID configuration, the RAID controller managed the disk and presented the operating system with a LUN composed of multiple disks.
The RAID configuration was stored on each individual disk and was re-established at startup time by the RAID controller after reading each disks configuration. This aids situations in which either the RAID controller has been replaced or disks have been moved within the system offline (referred to as "drive roaming").
The configuration on disk (COD) is kept in a consistent state by the RAID controller. In cases where the COD becomes inconsistent (usually due to a failing RAID controller or disk), the controller marks the drive as "dead" and attempts to bring a hot spare online.
Hot spares may be kept available to either a single array (a "local spare") or to the entire system (a "global spare"). These spares are allowed to be hot swapped throughout the system and will adjust their COD after roaming.
The Mylex controllers used in the TP9100 storage system provide for online expansion of existing LUNs (Mylex Online RAID Expansion, or MORE). During an expansion operation, the RAID controller(s) will continue to service incoming requests, thus the "Online" designation.
There are several restrictions when using MORE that the adminsitrator must be aware of:
- Between 1 and 6 drives may be added to a LUN at once, not exceeding 8 drives per LUN
- There must be at least 2 drives in the source LUN, but no more than 7.
- The new drives must be in standby mode and must not be part of another LUN
- The capacity of each new drive must be equal to or greater than the smallest drive in the source LUN
- The source LUN must be online during expansion
Share your opinion