samedi 13 mars 2010

Résumé du Pillar Summit par un journaliste américain

Pillar juices flash drive box - By Chris Mellor • Posted in Storage, 9th March 2010 21:18 GMT

Pillar Data Axiom storage arrays can go a whole lot faster, use less energy and be more reliable, thanks to a range of new features from flash drive enclosures to pre-emptive copies.

A Pillar Axiom array is composed of storage enclosures or Bricks, I/O controllers called Slammers, and a Pilot management component. The company is building a 2.6TB solid state drive (SSD) Brick, using thirteen STEC 200GB Mach8IOPS drives, said to offer 500,000 IOPS in total. The drives are laid out in a 12 + 1 configuration. A failed SSD can have its data rebuilt, with rebuild time possibly being less than 80 minutes. Applications can be expected to run up to twenty times faster on the SSD Bricks than on a SATA disk Brick.

Pillar says its "Roadrunner", an Axiom with the SDD Brick, differs from Oracle's Exadata 2 database machine because it is not just an SQL accelerator, providing very fast block I/O acceleration for any application, not just Oracle clients. Exadata 2 also relies on InfiniBand, whereas Roadrunner uses Fibre Channel. Exadata 2 is more scalable, though, ranging from 250,000 to one million IOPS.

Earlier this week, 3PAR introduced support for Mach8IOPS SSDs, and Pillar echoes the view that these are much more cost-effective than the ZeusIOPS drives used in EMC and many other vendors' storage arrays.

Pillar had previously adopted Intel SSDs. Product management director Gokul Sathiacama said that there was not much difference price-wise between Intel's X25-M and the Mach8IOPS product, but STEC had a faster time to market and so won the SSD Brick deal.

Pillar also announced a third-generation Slammer with support for 10GbitE and 8Gbit/s Fibre Channel. It has said that Fibre Channel over Ethernet (FCoE) is coming later this year.

It is announcing preemptive copy, in which the system tracks individual hard drive performance and detects potentially failing drives. The data on such a drive is then copied to other drives and the sysadmin alerted that a drive may be about to fail. Even if the drive fails before the copy completes, the RAID re-build will take less time. Preemptive copy is scheduled for an April delivery.

The company is introducing its own Axiom-to-Axiom replication feature that works alongside the existing heterogeneous replication, which is based on FalconStor and InMage products. It has been developed in-house, works asynchronously, and will be available in April. Application-consistent recovery will be added in a later release.

A federation concept is being added to the Axiom software suite. Today, an application could have several LUNs (logical units) each with their own Quality of Service, specifying things like performance and protection levels. These LUNs will be able to be federated and the whole set copied (replicated) between Axioms in April. "No one else has that granularity," marketing SVP Bob Maness said.

Pillar is also making migration inside an Axiom much easier through the idea of storage domains. This is a way of migrating data within an Axiom from older bricks to newer ones without disruption. Application data can be mapped to specific physical Bricks, specifying such things as Virtual Tape Library (VTL) deduplicating bricks, the SSD brick for high-performance data, and so forth.

What customers can then do is update their Axiom with new Bricks, ones using, for example, the SSD Brick in tandem with a 2TB SATA drive Brick instead of older 500GB drives. They then create a new storage domain, assign the new Bricks to it, then move the data in a few clicks from the old domain to the new one on the newer Bricks. Finally, they retire the old bricks. Storage domains will come towards the end of this year.

They are needed for the use of the coming sleepy or power-down hard drives, scheduled for an early 2011 arrival. In the Axiom, data is striped across drives in a Brick. Sathiacama said you need to be able to create a sleepy drive storage domain for backup or VTL use. You power that brick up (and all the drives within it), run the backup, and then send it back to sleep again.

Pillar is also intending to add a global Axiom management capability and reporting facilities to provide graphic trends analysis so that, for example, hot spots can be located and dealt with. Both will arrive in the second half of this year.

The SSD storage brick costs $135,000 list. Sathiacama said that using ZeusIOPS SSDs would have meant a price more than double that.

Pillar says it has around 515 customers with 1,250 Axiom arrays in production use. Apparently, when it was first founded and looking for a name, J Walter Thompson came up the name "TransFabrix", referring to the company offering both SAN and NAS storage. Pillar backer Larry Ellison contacted CEO Mike Workman and said he didn't like it, pointing out that TransFabrix sounded like a bunch of cross-dressers. Pillar employees were asked for alternatives and a wider focus group settled on one of their choices: Pillar Data. Thank you, Larry. ®

Aucun commentaire: