One of the questions I hear a lot from IT professionals is: Which flash storage option should I get? Essentially, the speakers agree that flash storage is the way out of a performance problem, but they can't determine which method is the right one for them. The choices of how and where to implement flash are overwhelming. My goal in this column is to help you sift through these options and pick the right one for your datacenter and your specific problem.
About half the IT pros who ask me the what-flash-to-use question have a very specific, pressing problem. I call this triage mode. Either one application in particular is giving them fits, or they have an environment that will no longer scale, such as VDI. The other half are in refresh mode -- planning their next storage upgrade.
Generally, the most pressing factor in the selection of flash storage is how much money you have. That shiny new all-flash array may be able to do millions of IOPS, but if you can't afford it, how fast it is doesn't matter at all. The other important factor is the kind of performance problem you have. For the purposes of this column, I will assume that you have clearly identified a storage performance problem.
How to make a decision in triage mode
If you have one application (say, a database) that is having a problem, and most of that problem tends to be read related, then a server-side SSD with caching software is the quick way to fix it. This is also true if you have a few host servers in a virtual environment that are having a performance problem.
If that application is clustered, or if more than a few applications or host servers are experiencing performance problems, then it is time to look at a shared flash system -- either a flash-assisted array or an all-flash array. Along with the shared flash storage option, a network upgrade should be strongly considered.
[Are flash hypervisors the answer to your I/O bottlenecks? See Flash Virtualization: Enabling The Software-Defined Datacenter.]
Unless it is time for a storage refresh, solving a widespread performance problem with a shared flash array versus a server-side solution is a comparison of costs and benefits. Server-side caching software has progressed in recent years. It can now handle write caching and virtual machine migration much more safely than it could in the pat. But the software and server-side flash hardware need to be bought, managed, and tuned individually. A shared flash array applies performance universally and, for the most part, uniformly. But the cost of a storage network upgrade may have to be factored into the overall cost.
How to make a decision in refresh mode
As a fanatical planner, I prefer working in refresh mode. In this mode, where you are planning your next storage architecture, I can't see how in today's environment you can consider anything but a flash power storage system. A fair amount of time should be spent on performance planning and on network design considerations. You have to know what your workloads require of storage over a period of time, not just when you think they are at their peak.
The other consideration is the storage network itself. In the past, most storage networks were upgraded because the previous generation was no longer available. You started buying 8 Gbit/s fibre because you could not get 4 Gbit/s. Both were fast enough for most hard-drive-based storage systems. Flash changed that; storage went from the slowest part of the infrastructure to the fastest. Now you can fully take advantage of 16 Gbit/s bandwidth, and a move to that storage architecture has to be strongly considered.
All-flash or hybrid array?
In refresh mode, there are obviously many other variables to consider. Probably the most significant is whether you should consider an all-flash array or go with a hybrid array. That discussion could fill up a book. I will take a closer look at that topic in my next column.
Solid state alone can't solve your volume and performance problem. Think scaleout, virtualization, and cloud. Find out more about the 2014 State of Enterprise Storage Survey results in the new issue of InformationWeek Tech Digest.
Bob Friday, Chief AI Officer for Juniper Networks, explains how the advanced technology is transforming operations.
Contact center leaders from 8x8, Awaken Intelligence, and 360insight discuss the importance of agent experience.
AI may force enterprises to rewire parts of their data centers so they are fully optimized to run such workloads. The question is do you use Ethernet or InfiniBand?
