Even best-in-class rebuild times expose data to hours of risk
Blocks and Files points to an Demartek study (sponsored by Pillar) showing that the Pillar Axiom 500 rebuild times are much shorter on high capacity arrays that similar EMC or NetApp systems.
The glaring data beyond Pillar’s performance, though, is the teeth-clenchingly long times that data is one drive failure away from catastrophic loss in every case.
The tests were conducted with about 50 500GB drives per system using RAID 5 (RAID 4 for NetApp), meaning the arrays can be rebuilt if one drive fails, but not two. So during the rebuilds of from 3 to 23 hours, if another drive fails, all data is lost.
Insert 1 TB drives into the equation, and your rebuild time (and vulnerability) doubles.
RAID 6 and other dual-failure protective schemes make this problem go away, but cost a little in capacity.
How are you dealing with this? I’ve heard that RAID 6 is gaining traction for 7200 rpm high capacity enterprise drives like Seagate’s Barracuda ES that are less reliable than 15K SAS enterprise drives (see Seagate’s 3.5″ Cheetah and 2.5″ Savvio for reference).
Does your RAID vary by drive class? What other magic do you apply to make this work?
Since the Pillar Axiom 500 stripes all volumes (LUN’s or FS) across a minimum of four and as many as 128 (5+1 disk groups) the risk of another drive failure in a single disk group causing data loss is greatly reduced. In additon, the Axiom gives the administrator a simple command to mirror the data internally on the Axiom in a RAID 51 configuration (RAID 5 stripe mirrored) resulting in the total elimination of the risk of losing data. Yes, it costs in additional capacity used, but so does RAID 6 and RAID 6 does not solve the issue of lengthy rebuild times (sometimes days) for high desity SATA drives. As stated above, Pillar Data has solved that issue and brings fibre channel drive reliability to SATA with it’s “self healing” drive technology standard in every Axiom.
Pillar also will only rebuild extents on the drive that are actually being used. The test was run without this technology. Those drive rebuild numbers are extreme worst case scenarios.
Also note that RAID 6 and RAID DP impacts performance and while it will “protect” your data against a double drive failure it doesn’t address the fact that during theses MASSIVE rebuild times the performance of the tested systems dropped dramatically. If you need absolute protection of data you can always create the LUN on the Pillar system with double protection which protects you from losing over 50% of your drives in that LUN.
I’d love to see an apples-to-apples comparison of dual-failure RAID schemes: storage efficiency, performance impact, rebuild time. Can anyone help?
We switched to Pillar from NetApp for all of these reasons plus the following:
1. Pillar has one-time cost for CIFS protocol and free FC protocol… with NetApp the SW pricing kills you as you try to scale.
2. The NetApp rep failed to show us a nifty StoreVault (NetApp product) because he did not make commission on that. Pillar is more honest and doesn’t withold information.
3. No single point of failure on Pillar and SAN comes clustered automatically… with NetApp this gets costly and for what? Slower performance and more data compromise
4. The NetApp sales guys are all ‘fat cats’ who get the job because of good track record but they spend their time milking Enterprise accounts and being political. Sometimes character counts… Compare to EMC who hires bulldogs out of Boston that massively create new accounts and beat NetApp by volume. Now the new EMC NAS is actually better than NetApp’s leading position NAS. At least Pillar got it right from the start.