What's HP got under wraps? Looks awfully flash and tape shaped

HP is indulging in a storage product-fest orgy at HP Discover in Las Vegas with an all-flash 3PAR, deduping VSA software, refreshed backup software, and a new tape library. StoreServ all-flash array HP is debuting the 3PAR StoreServe 7450, an all-flasher, with a tad more than half a million IOPS, 550,000 to be precise. We …

[Anonymous Coward]

Starting price of $99,000

I guess the starting price gets you two the array, the software and 2 SSDs, one of which is the hot spare. Or is it 3 - RAID-1 + hot spare?

Joking aside, great though this achievement may be, how many customers really need this amount of IOPS on a storage array? It's really a niche need, compared the vast amount of the world (small-midmarket) that runs on 1,000 - 25,000 IOPS across their entire infrastructure.

And for those higher-end customers - where do you need the IOPS - in the storage array - or better still in memory or tier-0 in the server itself with Fusion-IO cards or similar.

[Nate Amsden]

Re: Starting price of $99,000

It is somewhat of a niche of course(I'd imagine most of that is in the banking sector), but there certainly seems to be a market for an all flash offering with as HP calls them "tier 1 data services". The 7450 makes a lot of sense I think. The all-flash P10000 that was released last year made less sense to me(mainly due to power and floor space requirements). But HP told me that they did that all flash P10000 for customers who wanted it, they didn't care that it wasn't completely optimal, they wanted 3PAR and they wanted it to be all flash.

The 7450 just puts this in a much more efficient form factor, at a massive price difference as well. What took about two cabinets with the P10000 now takes 4U with the 7450 (the same size as a pair of P10000 controllers alone). Though the P10000 has massively more I/O ports than the 7xxx series. Cost I'd wager would be at least half of the cost of what a P10000 solution would cost - maybe less than half. All that and you don't take any compromises (relative to the 3PAR high end) on the software/features standpoint.

It'll certainly give EMC, NetApp, IBM and HDS a real run for their money, and put some(not as much as they put on other big guys) pressure on some of these new storage startups that specialize entirely in flash.

[Lusty]

Re: Starting price of $99,000

"Joking aside, great though this achievement may be, how many customers really need this amount of IOPS on a storage array? "

According to Violin, very few indeed. Although they were among the first to get this level of IO (and Violin can do it on random writes too) they concentrate on latency in order to boost application performance by reducing the time spent in a CPU wait state. 500k IOPS is no good if the latency is still 20ms.

[Anonymous Coward]

That's why HP are quoting sub 1ms latency for those numbers.

Providing very high performance coupled with low latency is relatively simple if your not doing all the other storage services. 3PAR is extremely feature rich and provides inbuilt high availability services the all flash vendors can only dream about and yet is still able to offer blistering performance.

The other upside is that this is managed like any other 3PAR array and can be federated with other 3PAR arrays whether all flash, all disk or hybrid.

[Lusty]

"That's why HP are quoting sub 1ms latency for those numbers."

Do you know the actual number? Violin are quoting latency in low us rather than ms so "sub 1ms" could still be orders of magnitude slower. Remember that a 3Ghz computer would consider waiting 1ms for something very boring indeed!

I'm sure the HP kit is fine, but from my perspective it is and has been for some time very slow to update with no real innovation being shown and updates/features only coming along when they are ridiculed for being out of date. When you look at the offerings from other vendors like NetApp, Violin, even Dell Equallogic it's hard to justify an HP box of any kind. You only need to investigate when the various vendors added support for VMware and Hyper-V integration to see who is actively developing and who gets around to it in the end. The 3Par may appear feature rich but that's no fault of HPs!

[Anonymous Coward]

3PAR doesn't want the 1% of the market that needs 1 Million + IOPs @ sub microsecond latency, Violin, Fusion I/O et all can have it, as to do that you have to rip all of the features out of the box and rely on host based software and mirroring of everything as Violin do. Also how many hosts can share that storage on the end of a PCI-E bus. Instead 3PAR they want to slot into the much larger high performance market with well established and robust feature set, that fits' within a broader family of products providing common management and federation.

Now you really are showing your ignorance, in terms of VMware support I think if you check your history you'll find HP 3PAR were first to market with a number of features and are either ahead or worst case on parity with the market leaders. Hyper-V the same with features like storage aware live migration between sites. Violin Has pretty much zero features that aren't supplied by host based software, Netapp are suffering from an identity crisis coupled with a severe lack of innovation on all fronts and Equalogic doesn't even pretend to be in the same class.

If you look where HP have taking 3PAR it's been continuous innovation over the two years they've had them in house. VClass 10,000 Series High end 2011, A-Class 7000 Series Mid range 2012, now 7450 all SSD 2013 along with many new features across the board and also now a $1Billion dollar run rate business. You really do need to do more research.

[Lusty]

I've done plenty of research thanks. HP since buying them have done what HP do best and alter the form factors, and redesign the hardware. They have also removed several of the availability rules which were making the 3Par too expensive and made them optional in order to sell to their usual lower end market.

Equallogic were first with both VMware VAAI and SMI-S support for Hyper-V. They also supported San copy offload in Windows long before Windows did. No, they are not in the same league for hardware but they certainly have better developers.

As to your first statement about 3Par not wanting that 1% - that's the whole point. They have added SSD because everyone else did and they don't appear to know why or who they are aiming it at. They have also done it by slapping in some SSD disks which is not necessarily the best way to go about it. Rarely in this industry does adding features to tick a marketing box work as well as designing a system around a requirement.

[Anonymous Coward]

Then your research is severely lacking and I'd suggest you book an update session with someone who understands the technology. BTW 3PAR had more R&D cash in their first year at HP than they'd had in their entire history. .

Comparing the development effort required for a small entry level system (no disrespect to Dell EQL here) against the development effort required for a solution that scales from low end (2 nodes + 8 disks) right the way through to high end enterprise (8 nodes + 1920 disks) implementations is hardly a good comparison, so these features on the two platforms are hardly comparable.

Besides the development complexities, testing and QA effort involved, 3PAR's firmware deployment model is designed to only release major features twice a year specifically to minimize the qualification and upgrade planning required by high end Customers.

At the low end you can have a feature release every other month and no one gives a toss, so being first by a month or two hardly makes for bragging rights.

[Anonymous Coward]

On the he availability rules you are also incorrect, the flexibility in those rules has always been in place prior to HP. By default the 3PAR will provide the highest level of availability possible with the disks available within the system by automatically protecting against the loss of an entire disk enclosure. If the Customer wants to override that policy because non of their other mid range arrays can or do provide that same level of redundancy then that's a Customer choice. Also keep in mind the availability rules are flexible, theyr'e not all on or off, they can be varied on a per volume basis so again it's about choice and the system will automatically maintain that availability through ongoing upgrades. You obviously know just enough to be dangerous. .

BTW IF you didn't already know you've been able to buy an all SSD version of 3PAR for a long time, so just adding SSD to 3PAR as your suggesting has happened here is hardly something new. The 7450 includes SSD specific optimizations as well as higher spec CPU's and Cache, the marketing way would be just to slap SSD into the current range and rebadge as many others have done, and that's just not the case.

[Lusty]

"We understand that it runs at 520,000 4K random read IOPS."

Who quotes random read IOPS?! Come on HP, lets have the useful figures or get back in your 90s shaped storage box where you belong.

[Anonymous Coward]

90's shaped box ?

Agree 100% read number is not what you'll see in the real world. But the 100% read number is also what all of the other vendors are also quoting. Actual performance wil be heavily dependent on the write ratio, SPC-1 numbers would help ther. Which HP have been happy to provide on the rest of the 3PAR family, unlike many other vendors, so maybe a little patience is in order.

[IO-IO]

yeah right

If you look at the SPC numbers for the all flash 3par it's nothing special. They are making a lot of noise about very little. The array doesn't treat flash specially, just cache algorithms were slightly changed, mainly to prevent this being a bottleneck.

You can build this system today on a 7400 - same hardware, it's just given a new model number so HP can have a dedicated flash product on it's books for analysts.

It's all very "me too" and frankly quite late..

[Anonymous Coward]

Re: yeah right

Actually most of the software changes are back ported to the 7400 and the rest of the 3PAR family, but not all, and also the hardware is different on the 7450 with upgraded CPU's and both data and control cache, So no you can't get the same performance from a 7400 as you would on a 7450 regardless of positioning.

"It's all very me too" really and that's why the 7400 general purpose array, not the new 7450 SSD array, blows everything else in it's class away before those software changes were even implemented in the new code.

[IO-IO]

Re: yeah right

The base spec for both 7450 and 7400 were the same. The 7200 has lower spec CPU's. There is a single code stream for the whole product lineup, every device gets the same features and capabilities. This has always been the case since the old S-class to current 10000 series.

Don't get me wrong, the 3par will be a good array that's trying to do the SSD-thing - way better than the rest of the mainstream vendors. The new kids on the block are just better, faster and often more cost-effective.

Go and test them. Few vendors will put their money where their mouth is. Let the tech talk for itself, not the product managers.

[Anonymous Coward]

sorry you're incorrect 8 core CPU's and double cache on 7450. Yes single code stream for 3PAR, but that doesn't mean an identical code path per model. Also S-class 32 bit, F, T, V etc all 64 bit.

Define better !

[Anonymous Coward]

Re: yeah right

Hehe, love your other post linked where you suggest Pure's secret to high performance.

"The real secret for treating flash well is to coalesce as many writes as possible."

http://forums.theregister.co.uk/forum/1/2013/04/09/ssds_and_raid/#c_1786154

That's pretty much what HP have done with the 3PAR arrays cache optimizations, so it should meet with you approval :-).

Oh and given you suggest ""It's all very "me too" and frankly quite late"".maybe you could share with us what EMC, HDS, Netapp and IBM are doing to optimize their hybrid arrays for flash.........er

[IO-IO]

Re: yeah right

There's only so much you can coalesce by holding off writes in cache. It was a minor tweak to cache code that applies to all systems running the current inform OS, Large disk based systems would also hit the cache bottlenecks under the right workloads. The 3par is a great disk array but its still largely treating SSD as disk. It's architecture is vastly superior to EMC and the rest of the rats in dealing with SSD but will not be as efficient as the purpose built next gen devices.

Don't forget when 3par came to be, they defined what a next gen array was, especially with Thin provisioning.

Technology moves on and 3par is now a mainstream technology trying to adapt to new concepts but it's still bound to an architecture that was made for disk. Put the systems side by side, you'll see what I mean :)

[Anonymous Coward]

not disputing architectural differences, show me the enterprise feature sets and availability on these flash startups.

The point being performance is easy if it's'you're only goal, once you add in high-end availability and features things get complex and code gets in the way of the data.

So if we take a holistic view of the arrays capabilities I'd happily put 3PAR side by side with one of these, and whist you're at it show me an independent audited benchmark for the same.

[Anonymous Coward]

That's not fast (per SSD)

I'm assuming their 550K IOps figure is based on a fully loaded system with 240 SSDs.

Which equates to 2291 IOps per SSD.

That's terrible performance.

[Anonymous Coward]

Re: That's not fast (per SSD)

No it's based on 48 x SSD's, so that's 11,458 IOPs per drive, or 5 times your estimate, do try and keep up.