Should You Adopt SSD Caching To Boost Data Centre Performance?

SSD Caching

The pressure to grow is relentless for data centres, but application speed is also an important consideration for administrators. By 2020 there will be 26 billion Internet connected devices, driven by the Internet of Things, according to Gartner. They’re going to be generating huge amounts of data that needs to be processed in real time. Is SSD the answer?

SSD technology offers superb performance. But, before we hail it as a panacea, there are things to consider. SSD’s are typically more expensive than the alternatives, and replacing existing systems will be enormously disruptive. What happens if the switch goes ahead and the expected performance boost doesn’t materialise?

It makes more sense to dip a toe in the water, than to jump straight in. SSD caching offers an opportunity to test the technology out, measure its efficacy, and make sure it’s going to meet expectations.

Challenges Of SSD Caching

A limited commitment to SSD with a new caching system is still going to present a number of hurdles to overcome. To install cache in the applications servers, there will inevitably be some downtime, as new hardware and software is implemented, and it may prove expensive. But you can mitigate these factors by opting for centralised caching with an appliance that can be inserted into a SAN environment. The idea is to make SSD available to all apps regardless of host, and to avoid the cost of completely replacing your existing storage system.

What To Look For In A Cache Appliance

However, even with a cache appliance, you really want to be able to plug into your SAN without a herculean effort. You also need solid metrics in place to provide reliable data on its effectiveness. Here are some practical points to keep in mind:

  • You want 5%-20% of SSD to be provided by your centralised cache appliance depending on your applications
  • You shouldn’t have to change LUN masking, zoning, application host configurations or anything else in the SAN environment.
  • You should be able to remove it, just as easily as you inserted it (in case it doesn’t deliver the benefits you expected or you want to move it to another SAN.
  • You must be able to fully analyse I/O traffic, covering individual paths, initiators, targets, and LUNs to identify bottlenecks, in case storage is not the real culprit for performance problems.
  • You need real-time access to data read/write patterns with intuitive explanations of cache hits and misses, and a detailed historical record on how the I/O access characteristics change minute by minute, or hour to hour, which provides precise information on how much cache is needed to optimise the SSD usage.
  • You should be able to select individual hosts or LUNs that are high priority, and suitable for caching.
  • You should be spending significantly less than the cost of switching all existing storage to SSD.

Take your time and shop around for the right cache appliance. If it doesn’t meet your needs out of the box, or it’s going to require a lot of disruptive tweaking, then it’s probably not going to be cost-effective or productively beneficial.

Taking The Measure Of SSD Caching

It’s very important to get your metrics right, so that you can compile detailed data on the cache and find out precisely how well it is working. There are so many variables in most data centers that it’s not easy to predict the impact. After the appliance is installed, it should be able to provide real-time, as well as historic performance data, even before cache is enabled. It may very well be that all you need is to balance the data paths, or break bottlenecks by redistributing specific LUN loads, you’ll be informed.

Trying SSD caching and measuring the results may provide the proof you need to justify a wider adoption of SSD. It could provide the stopgap you need to defer complete SSD adoption until you have the budget. Measuring will certainly help you fine-tune the system, so that you can extract maximum value from it. Strategic employment of the cache appliance may even deliver the results you need without a full SSD deployment, which could save you an enormous amount of time and money.

Whatever way you look at it, a centralised cache appliance is a sensible first step for any company considering the switch to SSD.

Paul Bjorndahl

Paul Bjorndahl is vice president of business development for Cirrus Data Solutions, a developer of Data Migration Server and Data Caching Server for storage area networks (SANs). He held sales positions at Small Tree Communications, Montillio, a startup company involved in Fibre Channel adapters, and Silicom, a provider of networking solutions. He began his sales career at Adaptec, helping to increase sales by over 50% per year for five consecutive years.