Primary Storage - 3Par 7450 vs Pure FA450

[EdwardPoll]

Morning!

A little off topic but I'm keep to get the input of real world users.....

We are currently out to tender for replacement of our spinning rust arrays, on the table we have all-flash-array offerings from HP (3Par 7450) and Pure Storage (FA450). Whilst we usually find there are clear differentiators that make our decisions on purchasing a reasonably easy task, this time round we're really struggling to make a decision. So my questions are, what do you all think? Have you compared them both? What are the pros/cons of both? With the underlying performance of the arrays, does Veeam snapshot integration really matter if you are reducing the risk of VM stun through raw power.....Once you remove the marketing and competitive mud slinging, which is the one to have!!?!?

Any light bulb moments greatly appreciated.....

Thanks

Ed

[dellock6]

Hi Edward,
first of all a note: I work at Veeam, and as such we are HP partners and as you said we have HP integration for 3Par in our software. That said, don't get what I'm writing as the Veeam official position, but my personal opinions.

First of all, let me give kudos to you for having data availability in your criteria of evaluation for the new storage array. It's not a common behaviour and sadly for some customers data availability is an after-thought once the environment is already built, instead of designing it from the beginning as an "organic" component of the datacenter.

Given the fact you have probably already evaluated the data services they have available (snapshots, thin provisioning, VAAI support for vmware, replication...), when it comes to data protection indeed the storage snapshots can be a great addition, but on the other side I cannot say how much the "raw power" of Pure can nullify the commit problems even without the storage integration, I've never tested one of their arrays (I'd love to ).
For sure, storage snapshots can help when the target storage is the bottleneck: in this scenario (and using an all-flash array at source the chances that the target will be the bottleneck are high) backups will take some time anyway to complete regardless the speed of the all-flash array. So, my first suggestion is to design your solution considering also this aspect and so properly size also the repository to take advantage of the new storage.

Second, I know for sure at least Pure gives the possibility to test their arrays in prospect environments, probably also HP can do the same. The best you can do is to do tests with real virtual machines their behaviour, instead of making choices based on documentation.

Last, I remember there are some other users here on the forums using both 3Par machines (not sure the 7450) and Pure machines. You can get real feedback from them. Have a look in other thread using our search.

Luca

[vStewed]

=== Disclaimer: Pure Storage Employee ===

Moving from disk to any AFA will result in a better experience for every application. Databases run faster, VDI is more responsive, etc. All AFAs are fast and likely will exceed your ability to push them to their limits, thus I advocate for the prospects to test / validate requirements in the areas of data reduction, resiliency and app/infrastructure integration before moving on to performance testing.

1. Data reduction: Technoliges like Thin provisioning, reduplication and compression reduce acquisition cost and slow the rate at which AFA capacity is consumed. All 3 are critical to having an AFA function as a general purpose storage device and not a point based solution. The devil is in the details with these technologies. Does thin provisioning include zero removal, pattern removal, and T-10 unmap? What is the level of granularity of dedupe: 512B, 4KB, 8KB or 16KB? The smaller the better as savings will be lost as server and VM file systems age (fragment) and data is challenged to be written into contiguous blocks. How many types of compression are in the system: one, two, three? The more that exist, the larger the return as different compression algorithms an encoding methods produce different results on different data types. Last, understand is data reduction is global or restricted to disk pools / groups. the former provides greater savings as it requires a smaller capacity footprint.

2. Resiliency: Most purchase AFAs for the consistent sub 1 ms latency - this is the attribute of flash that raises application performance levels. Is data reaction optional? If so, how much does it impact performance? Does the AFA support multiple simultaneous drive failures from the same RAID or protection group? Is there a performance impact to provide this level of resiliency? What is the performance impact when a controller is offline for maintenance or hardware failure? Remember AFAs will push more IOPs than a disk array and thus performance drops are felt with more pain than with a disk array.

3. Application & infrastructure integration: This area falls in to 2 buckets; what storage management capabilities are in the array and what extent into the data center ecosystem? Does the array provide snapshots, replication, clones, etc? Are they simple to use? What is the cost of these features? Does the array provide VAAI & a vCenter plug for VMware, VSS for Microsoft applications, and integrated with Veeam? What tuning is required for applications? Do I have to change the block size in the application? Do I have to load balance the data layout based on array controllers or RAID types? What AFA monitoring tools are included and what are available with additional licenses?

I think if you work through these three questions you'll be way ahead of the game. From there look at performance...

4. Performance: All AFAs are fast and are much better than disk - but often scale in performance in very different manners. So ask, what is the block size used to market the performance capabilities of the array? The larger the block size the better as applicants have multiple IO streams of virus sizes. This is well understood in the VMware space as the IO Blender effect. Does the array have a variable or fixed size backend? Variable will scale without excessive overhead or processing when compared to fixed. Like I mentioned earlier, whats the performance impact of RAID options, data reduction technologies and features like snapshots, replication, etc.

OK - I gave you a lot to go on here but these questions are much better than questions with implied value like 'whats the max IOPs?' or 'does the AFA scale up or out?'

Best of luck in your selection process and if I can help ping me. @vStewed on Twitter or virtualstorageguy.com

[HPEStorageGuy]

First, thanks to Anton for alerting me to the conversation. If it's not clear from my screen name, I work for HP Storage and have since 1990.

I'd start by pointing you to a document titled "Questions to consider when selecting an all-flash array". http://www8.hp.com/h20195/V2/GetDocumen ... c=us&lc=en. This is a really good document covering performance, resiliency, efficiency, and ecosystem.

Since you're already a Veeam customer, the integration with HP 3PAR and Veeam is unmatched. HP was the first vendor to integrate with Veeam - and by a long shot, going back to v 6.5. HP 3PAR snapshots have Veeam integration (Explorer for SAN Snapshots). You can do Veeam recovery from a storage snapshot, as well as a daily backup. That snapshot can be scheduled for every 5 minutes or as needed. Veeam also added snapshot integration (Backup from Storage Snapshots) to speed up backups. This snapshot integration can also mitigate “VM stun” on your messaging and DB servers during the VMware snap process, if that’s occurring. I have a blog post I did showing the integration that I just did at VeeamOn a few weeks ago. http://h30507.www3.hp.com/t5/Around-the ... a-p/174792

One of the claims we often hear startup AFA vendors make is that legacy storage retro-fitted with flash but not built for it. One of our engineers has a very good and detailed blog post discussing this and I can't recommend it enough. http://h30507.www3.hp.com/t5/Around-the ... a-p/175268. Don't miss the architecture discussions in this blog post. If you really want to dive deep, there is a technical white paper with much more detail on the architecture specific to flash: http://h20195.www2.hp.com/v2/GetDocumen ... c=us&lc=en

There are a number of things that set 3PAR apart: QoS with latency goals; 460TB raw scalability, Sync and Async replication (including 3 data center replication); ASIC-assisted inline deduplication (that never turns off), NDU SW & HW, 6-Nines availability (including a guarantee), VMware vMSC certified, under $2/GB usable. Note that HP 3PAR was also VMware's exclusive development platform for vVols, so integration with VMware is very deep.

Happy to answer or get answers to any questions you have. You can email me at hpstorageguy at hp.com or find me on Twitter as @HPStorageGuy

[EdwardPoll]

Thanks to all of you for your comments....helpful and thought provoking at the same time...

Pretty much all of the points that have been made we've come across and considered but and still we are struggling to find the single key differentiator..... hardware vs. software, software vs. hardware, variable block vs. fixed block dedupe, compression or not, Software operations bound to CPU vs. Software operations bound to ASIC, sync or async replication, Metro cluster or not, etc etc the list goes on.....all of these points have justifiable reasons to do them in the vendor chosen way but equally some have arguably good reasons not too...

In the end, I suppose its just down to the colour of the box...... orange or yellow........ and we don't mind either

Ed

[vmexpert]

struggling to find the single key differentiator..... of these points have justifiable reasons to do them in the vendor chosen way but equally some have arguably good reasons not too...

I know what you are saying here... every storage has its pros and cons, but in the bigger picture they are all largely the same.... so, Veeam B&R integration was that single key differentiator for us.

[vStewed]

EdwardPoll

I could give you a vendors biased comparison - or you can experience the difference by simply doing a proof of concept. Unfortunately for customers marketing messages and slide can confuse the difference between products that transform markets and those who try to slow sales by rebranding their existing products.

GO for the POC - your organization will thank you for it.

- cheers,
v

[emachabert]

Nice thread, great people replying here (Calvin, Vaughn...).

At least, whether you choose 3par or Pure you'll get the same shelves, just in different colors

From my experience and my point of view, both are just the best you can get at the moment.
3par has veeam integration, which is very nice and useful (eg point in time recovery from storage snapshots during working hours)