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.