Recently, Robin Harris over at Storage Mojo made a statement that caused a stir with both users and the industry. My summation of his column and subsequent defense of that column, is that he believes that the performance of a flash array is more important than the features that may come with that array. While I very often find myself agreeing with Robin, in this case I take a small exception.
The ROI of IOPS
In his column, Robin makes an excellent point about the value of high performance and how having that performance can reduce costs in other areas of the data center by better utilizing compute and network resources.
I agree with that position 100% but the key is the data center has the potential, at least some day, to tap into that high performance. If they can’t, then the storage system vendor is better off using the IOPS for features that will make the system easier to use and more cost effective.
My concern with the performance over features stance is that it may not hold water in the reality of the typical mid level to large sized data center. These data centers, thanks to the very features that are under attack here, like thin provisioning, snapshots, clones as well as deduplication and compression, can now afford to consider the use of an all-flash or flash-assisted storage system.
Features Come First
The reality is that, for today at least, the applications and environments in these data centers can’t generate anywhere near the performance that even the most burdened flash arrays can generate. For example, Pure Storage claims 200k+ IOPS per array today. Should a flash array be able to provide more? Sure. Do all data centers have an aggregate set of applications that can generate 200K IOPS? Nope.
Last year I did state that someday many data centers will need 2 million IOPS in performance, but today, or next year is not that day. There is however an increasing number of data centers where aggregate performance may indeed exceed more than 200K IOPS. For these data centers that are at that point or close to it already there are solutions from vendors like Violin, SolidFire and IBM. Each have flash solutions that exceed that performance plateau and do so with a full compliment of data services.
Many data centers can benefit from a flash storage system that has the performance they need and has features like deduplication and compression to make the technology affordable. Even the most feature burdened flash array will outperform its hard disk based brethren. For most data centers, making a modest sacrifice in performance that they wouldn’t be able to tap into is well worth the ease of use and cost reduction offered by these features.
How Should Features Be Added To All Flash
All this being said, there is a right way and a wrong way to add features to a flash array. There are two key developments that we see occurring to make sure this performance sacrifice stays modest, and we believe that eventually the performance sacrifice will be unnoticed by even the most demanding of applications. The first trend is a return to building purpose built flash hardware instead of using raw off the shelf components. This solves part of the performance trade-off by simply providing even more performance.
The second development is the improvement in the storage software that provides these features. Companies like Permabit are developing storage software that is truly multi-threaded and multi-core. This allows the processes that provide the above features, to leverage the advanced compute hardware that most of these flash storage systems come with. They can cost effectively scale storage software features to keep pace with advances in storage hardware performance.
Storage Swiss Take
The value of performance is often understated. As Robin details high performance can reduce costs throughout the data center. Increasing compute load, for example, can decrease the number of servers that need to be bought. But, the value of performance has to be weighed against an organization’s ability to tap into that performance.
If the flash array that an organization purchases today will more than exceed their performance expectations for the next five years, then they should focus on what capability and features the system has. Doing so will allow them to do their job easier and reduce the cost of the storage purchase itself.
Click Here To Sign Up For Our Newsletter
George Crump is the Chief Marketing Officer at VergeIO, the leader in Ultraconverged Infrastructure. Prior to VergeIO he was Chief Product Strategist at StorONE. Before assuming roles with innovative technology vendors, George spent almost 14 years as the founder and lead analyst at Storage Switzerland. In his spare time, he continues to write blogs on Storage Switzerland to educate IT professionals on all aspects of data center storage. He is the primary contributor to Storage Switzerland and is a heavily sought-after public speaker. With over 30 years of experience designing storage solutions for data centers across the US, he has seen the birth of such technologies as RAID, NAS, SAN, Virtualization, Cloud, and Enterprise Flash. Before founding Storage Switzerland, he was CTO at one of the nation's largest storage integrators, where he was in charge of technology testing, integration, and product selection.
StorageSwiss.com - The Home of Storage Switzerland 
George,
Thanks for joining the conversation. You and I agree that IOPS is not the important number for most data centers, which is why the first post was titled “IOPS is not the key number”. The point of that post was that in the context of all flash arrays the greater benefit comes from lower latency, not more IOPS, which, as you correctly point out, aren’t much use when the needed threshold value is achieved. So why, in your post, do you keep talking abut IOPS?
The second post “Data services more important than latency? Not!” was more controversial. I was responding to a Twitter thread – a first, I think, for StorageMojo – where an integrator CTO first asserted that customers don’t care about latency (true, but they should) and then questioned the datacenter savings due to flash performance. My response: where has this guy been for the last 10 years? How could he not notice the overwhelming vendor and customer adoption of flash and not wonder why?
This requires a longer exposition, but it points to a major blind spot in how we, as analysts, have looked at the market. Clearly we in the industry find latency a difficult topic, possibly because we’ve spent 50 years trying to make it go away.
Expect to see more about this on StorageMojo!
Regards,
Robin
Robin, I agree. Latency is a key issue. In fact a top read article of our is IOPS vs. latency but again only if the environment and the apps can take advantage of that lower latency. Whether it be IOPS or Latency at some point most data centers will reach a point where that can’t take advantage of improved performance (either IOPS or Latency) and features, ease of use and cost reduction, become more valuable. Looking forward to your additional details.
How can we (or the IT Admin) judge good latency? I see several flash vendors claiming sub-millisecond latency. Is that good? I think that is a bit long latency to achieve real-world IOPS at 1M. We see array vendors claim 1M IOPS but what is the usable real-world performance capabilities? How much hardware is required to achieve 1M IOPS? How much latency does their on-board storage features add? Is Software Defined Storage a better way to let a high performance storage controller do its job and generate low latency, high QoS IOPS while handling those array features at a different level? I think we may see a new architecture emerge shortly that can help answer these questions.
[…] it (next like: a couple meters away)! Let's start with Flash: See these two great articles about function and design of all-flash arrays (by George Crump of Storage Switzerland) vs. advantages of hybrid storage arrays (by HDS CTO Hu […]