EMC blows benchmark away - again EMC has blown another file-serving benchmark away with a result more than four times faster than the previous best. A pretty much all-flash VG8 (VNX gateway)/VNX 5700 array scored 661,951 operations per second on the SPECsfs2008 CIFS benchmark. The overall response time was 2.1msecs. SPECsfs2008 CIFS benchmark chart …
"The best non-EMC result was achieved by a NetApp FAS3210 with 64,292 ops/sec and a 1.50 response time."
Yes, but the NetApp benchmark only used 144 SATA drives and a PAMII module. Which is a considerable more realistic test than EMC's which used over 500 flash drives. What real value are EMC demonstrating by doing these tests? If they want to prove they are better than NetApp, why not do a like for like test?
These benchmarks need to be overhauled so that they show $/IOPS (or equivalent) and $/GB (end-user available, not raw) at a number of capacity points such as 1TB, 10TB and 100TB. And all costs should be over 5 years so that they include whatever insane year 4 and 5 maintenance prices these vendors think they can get away with. Only then will they be useful for end-users to compare one product against another in something approaching their own environments.
... they benchmark against another system which hasn't also been scaled for performance. This is no difference to fitting a VW Golf with a ferrari engine and then comparing it to a ford fiesta... remarkably enough the one with the ferrari engine won.
It's all bollocks anyway, as we all know, there isn't and won't be enough flash based storage available to replace spinny disks for many years to come.
Fail, because... well it's yet another piece of marketing crap, that bears no relevance in the real world, where people have to actually pay money for it.
http://www.spec.org/sfs2008/
Well - SPECsfs2008 is not a $/IOPS - It's performance benchmark and its pretty telling how a specific vendor solution can and will scale. When you (or I) develop an industry standard benchmark for $/IOPS against specific capacities, the benchmark would be no more real world than SPECsfs2008.... Just enjoy the fun and take it for what it is.
its 4 VNX systems with 8 File Systems (none of which can share data between them) hardly the kind of rig any business would deploy.
And for 6 million you get 60TB usable. Hell for that price point, I could deploy all DRAM and get twice the IOPS, faster writes and far greater reliability. Once again, marketing a system no one will ever buy as a valid benchmark.
SPEC continues to be a joke as well.
I love reading the posts on this page to keep up to date on storage news.
When I see posts like this showing one sided info based on configs that clients will never buy it frustrates me.
I agree show that EMC has a new fast bench mark. But put it into perspective that it is a config that is worth 5-6 million and that no one in their right mind would buy.
The trouble is, it may well all be a bit of fun. Vendors showing what their storage arrays are capable of etc. However, when EMC starts making claims that their storage out performs NetApp and pointing to the SPECsfs benchmarks it sets unrealistic expectations with some customers who don't know any better.
EMC's days of charging too much for storage are nearing an end.
Network wide global filesystems implemented in a redundant distributed way on Direct Attached SSDs will make these relics a thing of the past.
EMC's gear is irritatingly unreliable for the price and shamefully expensive with a byzantine software implementation.
VNX .. still a Celerra DART code at the backend. Yes. it scales from 1 to 8 data movers with independent back-end controller aka CLARiiON (dual-controller). But .. NONE of the DATA MOVER Share the SAME FILESYSTEM. 16TB limit ? It is still a Failover datamover concept with Control Station initiate the failover? 3 active and 1 standby? standby that sits there idle and waiting for 1 of the datamover to fail ???
I wonder if 1 use ZFS with similar number of SSD. I wonder if it deliver the same or better benchmark.
Seriously marketing hypes. But then again... it has been a long time since EMC deliver such hypes.
Yes it does show the scalability of the system. But then again, just pure NFS/CIFS without other real-world activities does not show the real system performance.
Whoever really use so many SSD ... you put a highly tuned Unix/Linux should deliver similar.
EMC only participates in benchmarks they can find a way to do well in. They are completely absent from the SPC-1 benchmark, where they would show very badly vs. 3PAR and HDS, among others. Amusingly enough, Netapp posted results for EMC Clariion a few years back, to prove this point (weeping and gnashing of teeth from EMC could not prevent the benchmark, check out SPC-1 results from Netapp to see the two results for EMC).
That said, I note with interest that on the NFS side of the sfs2008 benchmark, the "flash-only" EMC absurdity produces not quite 500,000 whatevers. IBM's Scale-Out NAS (SONAS), using spinny disks, produced just over 400,000 of the same whatevers (albeit with much higher latency towards the upper end of the performance curve). And HP produced a very respectable 333,000 whatevers out of a collection of HP-UX blades and a bunch of spinny disks. Where is HP's x9000 scale-out NAS box, and why post results for a non-commercial NAS config, when HP has a whole NAS line under the Storageworks division? We can be speculate.
Moving down the list, Blue-Arc both showed fairly pathetic numbers around 150,000 NFS whatevers for their top end box. Netapp only a little better at just under 200,000 whatevers at their high end, along with much lower numbers for their mid-tier offerings. And Isilon (seemingly a direct competitor for SONAS, and the leader in the scale-out space) only managed about 140,000 NFS whatevers. All very interesting, at least to me.
As to whether or not ZFS would show as good or better results if outfitted with a huge collection of SSDs, I don't see Sun/Oracle represented at all, which seems odd since Sun had the ZFS-based J9000 line before Oracle bought them, and even has some SPC-1 benchmark results posted (with pathetic IOPS figures, but at such a low price point that the $/IOPS were not too bad). They really should show up somewhere on this sfs2008 report, either NFS or CIFS or both... but, alas, N/A.
So, I guess I would like to see the graph that accompanied this story show both CIFS and NFS results, along with the various credible NAS vendors and all of their results, from either benchmark. I know the two benchmarks don't compare directly (CIFS "whatevers" don't equal NFS "whatevers"), but the NFS side does seem to be much more competitive, with at least one legitimate, commercially available spinny disk setup that almost matches EMC's best efforts at juking the benchmark.
The Register did a decent job of sorting the NFS side out here: http://www.theregister.co.uk/2011/02/23/enc_vnx_secsfs2008_benchmark/
Come on guys.. you can't criticize EMC because other companies submitted something more 'real world'. Who brings a practical family car to a horsepower war and brags about the affordability or the practical aspect? You know there are those out there that by the seemingly unaffordable high-spec no-real-world purpose cars.
There's nothing pro-EMC, blatantly or otherwise that I can see in this article. It's six short paragraphs that basically repeat what is says in the graph.
"EMC has blown another file-serving benchmark away with a result more than four times faster than the previous best."
True
"A pretty much all-flash VG8 (VNX gateway)/VNX 5700 array scored 661,951 operations per second on the SPECsfs2008 CIFS benchmark. The overall response time was 2.1msecs."
True
"The tested system had 560 x 200GB flash drives and 21 x 300GB, 15,000rpm SAS disk drives for a total of 101.2TB and eight file systems. The total exported capacity was 77.455TB. The VNX5700 had five X-blades, one of which was a stand-by blade."
True
etc etc
There's no analysis that says "the whole thing is too expensive to imagine and has no real world benefit" but surely we're clever enough to decide for ourselves what the implications of this new benchmark may or may not be.
EMC wants to create/reinforce the idea that EMC is 'fast' and doesn't care that customers will purchase spinning disk configurations that probably cannot deliever a tiny fraction of this level of performance.
I repeat: they don't care.
It's the same as base model and 'sport' model cars. One of them is the deliever high performance, in many cases well beyond what the owner/driver can extract and certainly beyond legal use of public roads, and the base model is to capitalize on the image of the car at a lower price point.
EMC wants to sell VNXs and it ideally would have customers who naively believe that they are 'better' and 'faster' than the competition. If you look at the SP utilization on a NS-480 or -960, I didn't think the VNX's massively more powerful processors was about crushing benchmarks with SSD. I think those powerful processors are completely necessary to be able to use thin provisioning, automatic storage tiering and, someday, block-level dedupe using primarily spinning disk.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
