HPE Storage Users Group

any updates on what roadmap for 3par..

seems kinda vague now with nimble

also is there any news of NVMe for 3PAR ? or gen 6 asic ?

as IBM has announce NVMe (Flashsystem 9100) , EMC PowerMax

Just taking a stab at this....

I will assume most people seeing a roadmap or getting a roadmap presentation will have had to sign a NDA so people that might know something are unlikely to talk.

My 0.02$...

Nimble and 3PAR is like comparing Lefthand and MSA. Totally different solutions primarily targeting different scenarios. Considering the push for Infosight I think that was a bigger reason for buying Nimble rather than the arrays which was probably more of a bonus. Considering the market share for 3PAR vs Nimble, I don’t see the focus shifting like it did with EVA vs 3PAR. I think both arrays will live side by side until they either sell off one of them (has that ever happened?), the external storage market dies and is replaced by software-defined storage or HPE buys something bigger/better..

As the 20k got a refresh and the 9k was released last year I would assume that the 8k(«low end») would be up next. Historically I think refreshes always include a new gen ASIC, but again... it might also be a 8kR2.... With gen4 the v-class/10k came before 7k, with gen5 the 8k came before the 20k... History doesn’t give any good hints, so I’m still guessing for 8k refresh. Any refresh would most likely be announced at Discover which is usually early December in Europe and mid-June in Vegas.

As for NVMe there was a lot of talk about 3D Cache/Intel 3D Xpoint a couple of years back with PCIe adapter... Not sure where that went but it kinda excludes current 8k as it only has 1 per node and in larger solutions the 8k needs the extra ports available there. The next question is where to put NVMe next.... today SAS can quite simply saturate all nodes in any storage system in an alarmingly low number of SSDs. I think and hope every storage vendor has NVMe on the drawing board, on a roadmap or in current products, but to be honest that isn’t where I hope all the focus is... We need faster CPUs and fabrics before we can start leveraging the true potential of NVMe... and remember, NVMe doesn’t do well with spinning media as it lacks all of the old SCSI commands for those media... So going NVMe backend = going All-Flash and flash only..... I don’t think we are quite there today(especially with current price of 3D Xpoint), but I think we will during the lifespan of the next generation of storage products...

_________________
The views and opinions expressed are my own and do not necessarily reflect those of my current or previous employers.

Exactly. I have signed a NDA myself, but I can say that both 3PAR and Nimble have focus in the roadmap I have been presented.

Pre-aquisition there was a fair amount of overlap particularly on the 8k arrays. Post acquisition we are seeing discounting models shift to push certain solutions.

At the end of they day they both work, Nimble is by far easier for non-primary support staff to support. On the flip-side the Nimble controllers are Active/Passive which alone is a show stopper for some. The Nimble AF's arrays have performed very well for us, on the flip side anytime a controller reboots it's a 27 second pause to all disk i/o which for some environments just can't happen.



If they are releasing new hardware ever, I don't see why not (even Nimble has an option). Remembering back when SSD's were first made available, pricing was crazy.

Long term, I think HPE is taking more of a wait and see what sells. If the majority of 3Par sales get consumed by Nimble...then i could see one direction. If sales stay strong, then another. HPE did by Nimble for Infosight but if the hardware also works and sells well...why axe it? Feature wise the Nimble's don't have the maturity that the 3Par's offer...if they offered active/active controllers and sync configs between arrays then this would be a different discussion.


Last go around, we looked at getting refreshes to our 3Par's but ended up going Nimble instead due to pricing. Overall they have worked very well outside of the Active/Passive controller design which has been less than ideal.

Nimble still has some catching up to do with 3PAR (some updates are planned) but the Infosight part was the main focus for HPE I think.

I would like to see the Xpoint NVMe card released but this is just another Read Cache and was only being talked about for the 20k (I guess it'll also come for the 9k).
I would also like to see 32Gb FC Cards, at a guess both of these could appear for current models (room permitting) but probably after another 3.3.1 MU release. However I know they are focusing on stability releases just now so that might be why new features haven't appeared recently.

I know they talked about a new asic a year or two back so it is on the cards but maybe they are looking at more NVMe features before release. I too would probably expect a new 8000 (although they blocked calling it a 9000 with the 20450 r2 marketing move ).

However any new major hardware design change will probably mean a new software release (3.3.2?) which could mean another 1-2 year of stability updates before I'd want to touch it for critical rolls.

My ideal system in 5 years time will be 1PB of NVMe Rulers in a 3-4U box (just so there's room for enough PCIe cards in the back and a ton of CPUs to drive it ), 32Gb FC NVMe (with added standardised LUN sharing support) and NVMe sync mirroring.

Although I'm not looking forward to the 100TB Ruler rebuild time.

Some quick thoughts on your question:

New hardware - I believe this is still under NDA so I won't get into details but there is a controller re-design to add more redundancy. This is one of the features I'm most excited about. All the storage SE's should have this in their slide deck so they can explain.

NVMe - as stated by others, 3dxpoint is their NVMe play right now. It should greatly lower read latency for certain workloads. Concerning an all NVMe array...I think they are far off on that. Apparently since 3par is a mesh of controllers with a mesh of SAS SSDs behind them, NVMe won't work in that way. In order to keep the enterprise features such as controller redundancy and shelf redundancy, the hardware hasn't been invented yet to do that with NVMe. You'll notice all the NVMe arrays being offered now are usually only 2 shelves with 2 controllers in each. Pure is like that. The EMC PowerMax is the same...but they make a grid of those powerblocks over infiniband to scale out. Unless 3par moves in that direction, I don't see it happening soon.

Good point.... SAS drives are dual ported so they can connect to two controllers... NVMe is "single ported" connecting to the PCIe bus.... How does Dell and IBM solve that with NVMe in their arrays? I'm guessing the controllers aren't sharing the PCIe bus

_________________
The views and opinions expressed are my own and do not necessarily reflect those of my current or previous employers.

again under NDA, most cant be dicussed.
the presentations i saw fall in line with how 3par has operated for a while now. not first to market with many options, but usually show up with a far less terrifying implementation then many other market competitors. they seem to be doing the same with NVMe.

as far as other features i saw were rather promising if linear progression of their current tech.

from the nimble roadmap..ill just echo what others have said. they are focusing on both, and providing some interesting delineation between the two product lines. i saw a well portrayed list of use cases moving forward for both.

thanks guys... for the updates...

if there are information that can be on a public domain, please do share it out..

fyi, the NVMe disk on EMC & IBM are dual ported nvme..



well that sucks, in this day and age.. downtime for systems is practically zero as required by business, then again, they always want things to be the cheapest...

Not sure if this entirely down to Nimble, keep in mind when you're reliant on the host to handle the failure, rather than something like persistent ports, then failover times can be heavily influence by the OS's own MPIO stack internal timers, which can often be tweaked.