IBM servers and storage to use Fusion-io currency
The two sides of Big Blue's SSD coin
IBM is likely to introduce flash-enabled servers and a productised QuickSilver within the next six months or so.
Flash-enabled servers will use direct-attached flash solid state drives (SSD) to accelerate their operations. IBM's Project QuickSilver had 4TB of flash SSD connected to 14 clustered SAN Volume Controllers (SVC) and producing one million IOPS. The SVC sits in a Fibre Channel SAN and virtualises, thin provisions, and protects the storage arrays connected to the SAN.
The QuickSilver flash was composed of one or two 250GB Fusion-io ioDrives with PCIe interfaces in 29 host servers connected by Fibre Channel to the 13 SVCs. There were 41 ioDrives in total.
IBM is working with Fusion-io to bring out servers with ioDrive SSDs, according to IBM Distinguished Engineer Clod Barrera. The company has been shipping SATA interface SSDs with its blade servers for some time.
It appears that IBM will also bring out mini-QuickSilvers, SVCs fitted with ioDrive flash. The SVC makes that flash sharable over the network, it adds protection via replication (synch and asynch) and snapshotting, and it provides a point of management.
We might conceive of a mini-QuickSilver product being built with two SVCs, to protect against a single SVC failure. These would link via a switched SAS backplane to individual ioDrives, each fitted with a PCIe-to-SAS bridge and each connecting to both SVCs for redundancy. These ioDrives could have 250GB, 512GB, or 1TB capacities, possibly 1.28TB as Fusion aims to reach that capacity point next year.
The miniQuickSilver product could scale out at the SVC level and also at the SSD level. Although it would be a sharable SSD storage facility each QuickSilver would treat its own SSD as direct-attached storage. Because SVCs can be interconnected in a cluster using Fibre Channel links to a SAN switch then a miniQuickSilver could be included in an existing SAN as an extra and very fast tier of storage with data migration between the tiers according to policies.
None of the technology involved is leading edge. It seems unlikely that IBM would choose to place SSDs directly in its storage arrays, the tack taken by EMC and Compellent. ®
SVC pretty darn stable...
"We might conceive of a mini-QuickSilver product being built with two SVCs, to protect against a single SVC failure."
So the redundancy already built into the SVC node pair isn't enough? It's pretty tough to kill a node pair, unless you REALLY try, or let the BOFH get involved...
Without ZFS, what do they hope to do with it? Better port ZFS to AIX!