EMC kills SPEC benchmark with all-flash VNX
EMC has topped the SPEC NFS benchmark rankings, scoring 497,632 operations/sec, using virtually all-flash VNX arrays. The previous top SPECsfs2008 NFS v3 score was 403,326 ops/sec from an IBM SONAS (Scale-Out NAS) system using 1,975 disk drives. There were 1,680 x 600GB and 240 x 450GB SAS hard drives and a total exported …
"436 x 200GB solid state drives (SSDs) and 21 x 300GB SAS hard disk drives, in four VNX5700 enclosures"
Your typical NAS configuration then.
A good example of how SPECsfs lacks a $/ops metric.
1st... too expensive to buy this performance.
2nd Not excellent utilization - RAW Capacity is 93.5TB. Exported is 60.243TB. 64% Utilization.
3rd .. Comparing to 15K Spindle drive ... mmm .. yes. Apple to Orange comparison. A Ferrari compare to Toyota Camry.
do admit it does give a glimpse of the storage controller max headroom. BUT WAIT! When we start to add those nice feature .. FAST V2, Thin Provisioning, Snapview and etc ... How much overhead and IMPACT to the system since all these are software running on top of the VNX system. If running thin provisioning can deliver same or equal performance this will be a joy for all Customer.
Time and Time again EMC creates useless benchmarks to confuse customers. Where on earth one customer can have 4 SSD backend arrays for achieving just 500,000 IOPS. Only EMC will be it's own customer with this config. Both BlueArc and NetApp have been honest in benchmarking.
Can EMC provide simple benchmarks for NFS with 100 SAS disks on 1 VNX? Going by their FAST ratio of 5% SSD and 20% SAS and 75% SATA, Pretty sure cannot cross 50,000 IOPS.
I would prefer they do another benchmark with 8 VMAX behind and cross Million IOPS.
I agree with AC "Usual mislead of customers #".... This is like saying you can put four totally different arrays side-by-side and count the performance on all four of them.
Bluearc/Netapp or whoever can just multiply w/e benchmarks they have by 4 and it would easily compare w/ this and blow its brains out even w/ the SSDs
Look at the file systems exported 8.... it is like having 8 different boxes....
To make the correct comparison of 8 exported FS. you have to multiple the Netapp number by 4 to compare it correctly. So, ~190*4 would be 760 w/ SAS drives only imagine what ti would be w/ SSDs.
BlueArc and NetApp does nearly 200000 IOPS without SSD on a single array. You multiply by 4 what happens... They become 800,000 IOPS. Checkout http://www.spec.org/sfs2008/results/sfs2008nfs.HTML for all vendor results to understand things better. What is filesystem got to with drives? I am sure EMC Isilon can fare better here but your argument is baseless Also check on the same link, historically EMC uses 2 VMAX or 2DMX in the backend and now 4VNX. Also think how customers can afford 4 times the price for 500,000 IOPS from EMC with or without SSD. Think practical in customer relevant terms is my request.
I'm so very happy that EMC has done this... maybe it will hasten the demise of all these stupid storage benchmarks. SPEC, SPC, etc all are completely useless and meant only for vendors to see you something. Now all the other vendors will go play the same stupid game to have the top numbers. Making these already useless benchmarks more obviously useless to the masses.
Hopefully all customers will then start throwing it back at the vendors like I've been doing for years now calling them on how it's complete and total BS (you won't believe the argument I got into with a Sun/Oracle sales guy). The only valid benchmark is one done to simulate my workload. If the vendor can't benchmark my load, then sell me a solution with a SLA to support a stated IO workload; if they are confident in the solution they will stand behind it (I haven't had a vendor walk away from a deal yet requiring a SLA with a stated IO workload).
All customers need to make vendors to do Proof of concepts and performance tests in their environment. All these useless benchmarks will stop.
Where the rubber meets the road here is that at the end of the day we might as well be throwing bones and reading tea leaves to understand how a particular workload will behave on just about any storage architecture. The only way you *really* know is seeing it, but even then, I've seen more instances than I care to admit with shared storage solutions working really great... until you load it up and share it the way it's intended to be used.
The software vendors usually don't know or don't care to define in any reliable method what the storage performance requirements are, and on top of it you get crap like this - "[yes, with our SAN latency will be less than the speed of light *AND* it will make sure your shoes never go untied again]" - and the only way to ever really know how a particular architecture will behave is by seeing it in action.
I wonder why EMC felt it needed to fill its VNX array with solid state drives and couldn't use one of its sub-$10,000 VNX arrays that it touted so highly in its marketing pitch last month? An all-flash array would be far too expensive for anyone except Fortune 1000 enterprises. This was a nothing more than a publicity stunt.
For years EMC was the only T1 vendor who stated that the SPEC benchmark was pointless, proves nothing and is a waste of everyone's time and money as it told you nothing about real life configs or work loads. They were widely criticized for trying to hid something. They only recently joined and guess what? Smashed the record with an unrealistic config on a generic workload. I think they have deliberately set out to prove the point they have been making for years here. So for all you morons on here having a go at EMC - if you knew your history you would wind your neck in. Don't hate the player hate the game...??
Sign up, sign up for Blocks and Files, The Reg's weekly storage newsletter
