
Meh.
Benchmark, schmenchmark. Featureset, resilience and reliability, support services and price are usually more important than artificial performance figures.
A HP 3PAR system is top of the storage heap in the SPC-1 benchmarks for storage systems costing less than $200,000 and also achieved a top price/performance ranking. Of course, many of the big boys failed to make a showing, so we're not sure how valuable the crown is. Out of all systems submitted to the Storage Performance …
"Matt, but you also know 3PAR has those too." Yes, I do, but do I need every feature in every situation? We recently had a need for external disk on a cluster of remote fileservers, and a reseller rep immediately peddled out the 3PAR 7400 rah-rah-rah routine. Nice kit, wonderful promise of oodles of IOPS, and a nice price even, but total overkill for the requirement. No need for 100+TB capability, all we needed was an additional 4TB useable. No need for remote replication, no need for thin provisioning or virtual copies, all already covered by software on the servers. What we needed for that particular requirement wasn't a kick-ass array but a pair of JBODs. I'm sure there are plenty of siuations where the 7400 is the best option, but I wouldn't make that decision based just on an IOPS benchmark figure.
I don't think anyone was suggesting you would, If it's over kill and you want some JBOD then that's fine too. I think some of the channel and the competition forget HP has other options. Maybe it dilutes market share to a certain extent but it does mean not every problem looks like a nail ;-)
"I don't think anyone was suggesting you would, If it's over kill and you want some JBOD....." Well, we do still have plenty of EVAs which will probably be replaced with 7400s at some point. :)
"..... I think some of the channel and the competition forget HP has other options....." They had a similar keenness for the P4000 kit as a "solution for all your ills" a year or so ago, I put it down to where the hp marketing money is being plugged from year to year.
"Latest is '12 for clustered FAS6240. Not so old"
http://www.theregister.co.uk/2012/06/26/netapp_cluster_mode_spc_benchmark/
And not so fast given this was their high end box running cluster mode, complete with 6 nodes, 288GB Cache, 3TB's of FlashCache and 432 SAS disks.
Oh and how hey crowed about the latency :-)
Ahh yes if you want speed and size then take a look at the previous 3PAR 10800 results
http://www.storageperformance.org/benchmark_results_files/SPC-1/HP/A00109_HP_P10000-3PAR-V800/a00109_HP-P10000-3PAR-V800_SPC1_executive-summary.pdf
That's what the disclosure document is about so you can judge is this an all SSD config, an all disk or a hybrid configuration. In this case it's stacked against all SSD systems from HDS and IBM and beat both by a wide margin.
The SSD based tests are really to show the controller limits of the system, not a particular capacity point, if you know where the controllers bottleneck then you know where you can scale to in the real world. In this case the test shows you can't bottleneck a 3PAR 7400 with spinning disk, whereas you can the others, so 3PAR has performance to spare which provides future proofing.
You might want to check this out, the comments are pretty revealing around the use of sparing on the IBM & HDS configurations (page 12 on each report). http://www.techopsguys.com/2013/05/23/3par-7400-ssd-spc-1/
I'm not sure you can say this was particularly poorly attended, after all the two people missing were just the usual suspects. EMC who won't provide any public benchmark where a price needs to be attached and will only provide internal fictional metrics under cover of NDA. Then there's DELL who's first, last and only submission to SPC-1 was for a PERC controller with 56 x 18GB SCSI drives back in mid 2002.
There's nothing to stop either of these two submitting results, both EMC and Dell are SPC members but in the case of EMC they have, rather oddly, never submitted a single result.
http://www.storageperformance.org/about/roster/
It's not obvious at first glance (left hand side, half way down the page, no logo and very discrete small print :-), however continued membership does allow them to veto anyone else attempting to publish results on their behalf, like Netapp did that time (which ultimately backfired and played into EMC's hands).
It's a benchmark, so all fun and games and it'll likely be beaten again in a few weeks with some crazy configuration, but the feeble excuses for not taking part have now worn very thin.
Just checking the reports and I see Huawei with a dollars per IOPS figure of $0.57 (not the 0.98 quoted above) which seems better than that of HP ($0.58) and a total IOPS of 400,587.11, also better than HP. It's also interesting that HP have gone down the more/smaller line for their SSDs (32x100GB) compared to the other vendors - Huawei (32x200GB), IBM (18x200GB), HDS (20x200GB). Special sauce or just more drives? How does the HP box scale?
In the end it just seems to be back to "Lies, damned lies, and statistics".
The point is flexibility the Huawei box turned in excellent results but like all the other All-SSD arrays, it's a one trick pony. The 3PAR is a general purpose array, you can load it up all SSD, all Disk or Both. In addition it has a ton of availability and data services built in that just aren't available on those boxes.
The SSD tests are all about balance and bottlenecks, a vendor will only put sufficient drives in the box to the point they bottleneck the system, adding more just increases the $ per IOp.
Not really, the IBM V7000 competes for the same market a HP 3PAR 7400, If you wanted to compare the 3PAR 7200 then it would be against the IBM V3700, I think we could probably guess the result.
All use a similar canister http://www.sbbwg.org/home/ but both controllers and software are completely different between IBM Vx000 and HP 3PAR.
My advice - even if you just love one storage vendor's kit over all others, ALWAYS have at least a look at the competition, if only to keep your incumbent vendor on his toes. It never hurts to have a trial bit of kit from vendor B "accidentally" in sight when the salesgrunt from vendor A comes calling. And never EVER tie your stack so tightly to one storage vendor's featureset that you couldn't (even if painfully) rip it out and replace it with another storage vendor's kit, otherwise they know they have you over a barrel. Storage is becoming commodity so treat it that way.
And if they say "we have the top score in benchmark X" just remember that your business is highly unlikely to be running benchmark X, it is running real applications with real users and real data in a real environment, and it is how the kit performs in THAT environment that is the only real benchmark that matters. If in doubt make the vendor or the VAR pay for a proof of concept. Benchmarks are at best only a vague indicator of how much marketing money the vendor has invested in the product. A proof of concept - especially a competitive one - demonstrates not just the kit's perfromance but also makes the vendor and/or VAR perfrom so you can judge them too.
/What, another team meeting down the local? On a sunny day? What a coincidence!
Disclaimer: I work for IBM and was involved with the V7000 all SSD publish.
1. V7000 like 3Par is clusterable.
V7000 publish used a single system, but can linearly scale to 4 control enclosures
So 4x the 120K published number.
3Par used essentially equivalent to 2 control enclosures, hence double the performance.
2. V7000 = 18 drives
3Par - 32 drives
3. SPC-1 tests benefit from cache to usable capacity of storage ratio. Hence with the 1TB usable and 64GB cache in 3Par - much higher cache hit chance than V7000 with 3TB usable and 16GB cache...
If done apples to apples, I thnk you'd find they are both limited by the SSD performance and not the controllers.
As with all things performance - devil is in the detail.!
No disrespect but clustering on the V7000 isn't really the same is it ?
What you really end up with is two ALUA failover clusters with each node pair owning their own storage and each node owning it's own LUN. Which in turn means you can't stripe the volumes over all 4 controllers, whereas the 3PAR 7400 provides fully symmetric active active clustering and wide striping across the entire array and all controllers. Also I doubt that buying and strapping together two separate V7000's associated licensing and support would stack up commercially vs a single system like the 7400.
As for the controllers not being the bottleneck, that's bogus, with SPC-1 economies of scale come into play and directly affect the $ per IOP. If you could get significantly more IOps out of the system by adding SSD then you most certainly would.