HPE Storage Users Group

We are currently comparing a new 3Par 7400 (4 Nodes and about 100TB) to our existing EMC VNX/VMAX environment.
Anyone have any disaster stories? i've read thru most of the documentation and everything seems pretty good.

Of course HP has promises, but how is support? how are the upgrades? does support assist with the upgrades?
How is expansion? do you require HP to assist or is it as simple as popping in some extra disks?

Anyone have any failures of the entire controller pair (my biggest fear)?

Any big performance issues? we plan on doing either all flash or a flash / 10k mix with AO.

We are pretty much 100% virtual and this seems like a better solution for us.

thanks for any feedback here.

I will type a full reply later, but I wanted to answer your reliability concerns...

On the 4 node and up, hp actually goes so far as to guarantee six nines (99.9999) up time.

I have 5 7200's, 2 of which have been in service since mid 2013, and we have yet to see a moment of downtime, planned or otherwise. We are currently at 100% uptime

To answer the parts on upgrades, this is something hp assist with, if it's a software/firmware update then it's covered by you hp support contract, if it's additionsl disks or shelves you are purchasing then you need to purchase installation services as part of the kit list.

It's a very reliable system as mentioned above, if it ps the 7400 you are looking at then that can have 4 controllers.

Make sure you understand AO before going for it, lots of people don't like it, as it's very slow to react, it moves data based on hours and not seconds or minutes, so a lot of people feel it doesn't fit their needs. By the time it's worked out that the data needs to be moved it's too late etc. it also has a tendency to move everything down also.

A more relevant feature might be to use SSDs as adaptive flash cache instead in the spinning disk cpg and then use dynamic optimization to move between the two tiers etc.

It carries a lot of performance just with spinning disk though because of the asic etc.

We have been HP 3par for over 3.5 years and now have 6 production arrays (V400s, 7400s and 7200s) and about to purchase 2 7440s. I can tell you overall the product is great. The technology is solid and these things are just beasts for performance.

The way it spreads data across all the chunklets across all controllers where no volume is owned solely by a single controller is something most other products don't do and that is what hurts them.

HP support has been a true roller coaster ride, crap shoot and we have mission critical support with a TAM. Once a problem gets to L3 or engineering it is like the black hole.

In theory there is a lot of redundancy int he array and we have had a controller go down without any interruption to service, but there is a single weakness that under correct circumstance can hurt you. Each controller has it's OS on a single SSD drive and if that drive has issues, but does not go down then you can have a scenario where that controller bogs down the other controllers. We had that happen once and our V400 went from serving 70K+ IOPS to under 1000 IOPS as the controller brought the array to its knees. This is a rare scenario and we have had the internal SSD drives fail on other controllers without hurting the array.

Overall even with a few negatives (every product vendor has them) I am still a huge 3PAR fan.

All platforms have failures - software / hardware & operational, I could point you to a few outages (documented online) involving EMC platforms (big and small). No vendor is immune to this so asking the question in that fashion won't always get you an accurate response, it's how the system copes with failures that matters e.g. Did they lead to downtime / data unavailability / data loss etc. Where extreme availability (regardless of the failure type) needs to be absolutely guaranteed that's typically where a DR/BC solution gets mandated by many organizations.

Support depends on what you're looking for and the level of comfort required HP have 3 main support levels :-

Foundation Care is a reactive service, so think break fix with dial home - probably not recommended for something as critical as a primary storage array.
http://www8.hp.com/uk/en/business-services/it-services.html?compURI=1198271

Proactive Care builds on foundation care by adding an optional ASM and additional proactive services around firmware etc. - This is the typical level sold with HP 3PAR
http://www8.hp.com/uk/en/business-services/it-services.html?compURI=1079424

Critical Support builds on both of the above and allows for Call to Repair guarantees, personalised services and much higher levels of service customization
http://www8.hp.com/uk/en/business-services/it-services.html?compURI=1079443

Additional adhoc services can also be added to any of the above through credits.
http://www8.hp.com/uk/en/business-services/it-services.html?compURI=1079391

Yes HP does assist with the upgrades both planning and execution, firmware updates are online and included in all of the above support levels but you can also choose to apply these yourself once familiar with the process. Dependent on the platform i.e 7000 series then yes this can be as simple as popping some disks into the system and hitting the tunesys button, but typically the hardware is sold with installation services which come at a minimum cost anyway so it's your choice.

Why would a HP 3PAR be more likely to lose an entire node pair than say EMC would be to lose a controller pair on a VNX or VMAX lose an entire engine ? Each node runs it's own local resources and O/S and the design center of the system pushes a dual node failure well beyond a 6 nines requirement anyway. However if you absolutely need to protect against such failures (regardless of the odds) on any platform from any vendor, then you need a second array with replication.

All flash is very fast but the way 3PAR works means you could quite easily see stellar performance on a combo of SSD & spinning disk compared to what you're running today. If you go the hybrid route then AO is a good solution in a mixed SSD and 10K environment and you can combine this with Adaptive Flash Cache to catch the outliers between AO runs. The issues mentioned above with AO tend to relate to where Nearline drives predominate, which due to their very low IOps/GB is a perennial issue for all vendors / platforms. If you go 7440 then you get the best of both worlds, all flash performance at the controller level but the flexibility to add spinning disk.

As mentioned above if 5 nines and the experience shared here isn't good enough then go with the 6 nines guarantee. However you will need minimum support levels and services to achieve this guarantee as you would were you to expect the same guarantee (assuming you could get one) from any other vendor.

For some other user opinions take a look here
http://www.itcentralstation.com/products/3par
https://youtu.be/wtZOD2eyVQc

Edited for clarity.

I installed our first 3PAR T800 in early 2009, it is still running in production to this day, with 100.00% uptime. We've added shelves/drives on multiple occasions, added FC adapters for remote copy, updated the microcode/firmware 3~5 times, replaced failed node, hard drives, hard drive in nodes, ram, batteries, power supplies, components of the shelf called FCAL, as well as replaced/upgraded service processors... all of which, was done hot and online by 3PAR and HP field support.

Support has been for the most part automatic. The service processor phones home, errors are remotely evaluated, parts and tech are dispatched... we get called to schedule letting the guy in the door to do his work. During the transition from 3PAR to HP, there were some rough patches with support. Explanations were given that sales and adoption of 3PAR skyrocketed and the support infrastructure was lagging behind. That seems to be behind them, and every thing is "up-to-par" now... (couldn't resist, sorry).

Prior to 3PAR I was an EMC guy (customer), Symmetrix and Clarrion were my go-to solutions. The 3PAR design for wide striping and chunklets was the end of disk based raid groups in our shop.

_________________
Richard Siemers
The views and opinions expressed are my own and do not necessarily reflect those of my employer.

I work for a service provider and we got 30+ systems of all models now.

We do not keep statistics, but we are definitely four nines plus on uptime,
I would not be surprised of we are close to the six nines HP marchitecure dept says.

AFAIK all outages has been during upgrades. We have probably performed more than 100 upgrades and expansions,
most without issues, some with minor issues and 3 outages, of which one where our own fault.
HP does all upgrades and most expansions.
All our disk failures (less one, see below) has been predictive and disks has been automatically evacuated pre-failure.


The only gripe I have is that QA tends to fail at times.
Dont touch any patches for the first two weeks,
wait at least for one or two maintenance releases before going to a new release.

We were hit with the bad disk FW for 3TB disks - lost 300 TB backup. All restored from tape though.
Got hit with the failover policy change in 3.1.3 all LUNs went RO during upgrade of second system,
as luck would have it, pre-production environment.
HP was not aware of this change themselves...

The last major outage was caused by a failed SSD drive on the controller during upgrade.
This had initially zero impact as this was a 4 node 7400. (a big advantage 3PAR vs comparable systems)
HP would come back on a Saturday to replace SSD and upgrade last node.
HP only tried once to contact weekend shift, and logged no respose from customer.
Assuming the upgrade was done, one of our engineers started up the controller using old FW version and brough everything down.
Entirly our own failt, but process and failsafe should be improved.
I belive 3PAR will evict a controller booted with old FW now?

We have requested redundancy for controller SSD and I believe this might be standard in future high end systems.

We got 3 AFA now, and I cannot say how happy I am with it.
Just go with the high capacity 2TB SSD unless you have really high performance requirements.
AO, AFC etc are just band aid to solve the old spinning disk latency issue.

I do not think there are systems as easy to expand as 3PAR,
add some shelves, run tuneys, and you are done.

Wide striping is a double edge sword though, a single server going haywire could impact everything else on the array.
Hence the SSD only approach, all latency, external workload issues and customer complaints are gone.

Compared with VMAX with all its pecularities and limitations, managing 3PAR is a breeze.
(We got a 40k and it is a beast, but really not userfriendly to say the least)

Just to add a little to the wide striping comment, if you want to tame that wayward server workload there's also 3PAR Priority Optimization (storage QOS). Handles bandwidth + IOps caps, minimums and latency goals as well as relative priorities and is enforced in real time. Works for all media types and will become over time just as important for SSD based workloads.

Yes, another great feature on the smorgasbord of optional features costing top $$.

I wouldn't assume QOS is only required for wide striped arrays, many times I've seen a misbehaving application or a badly written query with access to just a few disks worth of bandwidth cripple controllers on traditional systems due to large sequential transfers.

Yes it costs, but now part of the Optimization Suite V2 so you get at least one of the included features for free versus buying them individually It's not like the rest of the market don't charge for these kind of features in one way or another, even the all inclusive boys are taking their cut somewhere along the line (usually bundled within the appliance model) and if they don't you'll find features stagnate due to lack of development funding.

Many also don't have anywhere near the functionality of Priority Optimization, yes they can tick the QOS box, but dig in and you'll find multiple limitations that mean the feature has limited use cases.

You also have the option of QOS at the fabric or hypervisor but again you still pay for the feature one way or the other.