The good news is that users will likely see a substantial performance boost no matter which choice they make, according to Dennis Martin, president at Demartek LLC, an Arvada, Colo.-based industry analyst firm that operates an on-site test lab.
You can read the transcript or listen to the MP3 file found below.
SearchStorage.com: You use solid-state storage in your test lab. What sort of results have you seen? Any surprises?
Martin: I guess the first surprise, if you can call it that, is that we didn't need a lot of solid-state storage to get a really nice bump in performance. We've tested disk drive form factor [and] we've tested other kinds. In general, you get a very nice bump in performance. Things are faster. And of course, you have the benefits of lower power, lower heat, low noise, that sort of thing. So we like 'em, and I'd tell people to go for it.
SearchStorage.com: IT shops interested in using solid-state storage have a number of options. Can you outline the major decision points they face?
Martin: The first option with SSD is to decide whether you want to deploy it in a caching sort of configuration or more in what I would call a primary storage configuration. [A caching configuration involves] letting the controller or the disk array or whatever the host is for the solid-state do caching and let it determine where the hot spots are and then make the best choices for hot data; [in a primary storage configuration, you need to decide whether] to put specific things directly on there.
Read the entire report on solid-state storage
Solid-state storage technology: App needs dictate choice of arrays, cache, appliances or servers
SSD adoption slow despite rising enterprise solid-state drive shipment
Then the second decision point to face is, of course, what kind of form factor are you looking at? Are you looking at a disk drive form factor? Are you looking at something on a PCI Express card?
SearchStorage.com: In which scenarios are solid-state drives in cache a better choice than solid-state drives in storage arrays?
Martin: You would choose caching if you have data that's variable as far as the hot data concern. It's not always the same data that's hot, but it's a lot of different data that's hot sometimes and not hot other times. Or, if you have a wide variety of data that you want to accelerate -- if it's not just from one application, but from multiple applications -- you might go the caching route.
If you decide [on] what I call the primary storage [route] -- or put the drives directly in and place the data on them yourself -- that works better if you have very specific data that you know you want to accelerate, and you just want to focus on that.
SearchStorage.com: You mentioned form factors of SSDs. What sorts of things do people need to think about when deciding between PCI Express and hard disk drive form factors?
Martin: The first thing is the overall amount of capacity. If you're going with PCI Express, then of course you're limited by the number of slots in your server. So, although you can get quite a bit of storage on a single card -- I've seen 1.2 TB on a card -- if you only have five slots, then the most you can put in there is [approximately] 5 TB, for example. In the disk drive form factor, you can fill racks and racks full of disk drives, in theory at least, with SSDs. So, you've got a lot more headroom for capacity in the disk drive form factor.
SearchStorage.com: What factors should an IT shop weigh when deciding between solid-state storage in arrays, appliances or servers?
Martin: Of course, cost will be the key with all of these. It also depends on how much you want to do and where you want to do it. You can go with solid-state in the arrays; that works well. You can replace some number of spinning disk drives with solid-state drives. As I said before, it doesn't need a lot of capacity to do that. If you do it in an appliance, then you have the option of having that solid-state storage accelerate multiple disk arrays behind it, for example, whereas if you're in a single array, then it only accelerates the data in that array. If you put it in the server, then of course you're going to get some benefit across multiple storage targets because now the server has it inside of itself. It depends on what you want to accelerate.
SearchStorage.com: What should an IT shop do to determine which applications or which workloads will see the greatest performance boost from solid-state storage?
Martin: The first thing you have to do is measure what you're getting right now. You need to measure your raw performance -- IOPS and megabytes per second. You also need to be aware of the latency because some applications are a little bit more sensitive to latency.
Assuming you've done that measurement, then you can say, "Here's an application we want to use to put in solid-state." For example, you could put a database index in solid-state storage, and you'll get a very nice performance bump just by putting the index in there because the index is accessed relatively frequently and isn't terribly large. Some other applications really benefit if you put the entire application on solid-state. If you only put part of it, then you'll get a minor bump in performance. The key thing is to have measured what you're already getting and know where your pain points are, where those bottlenecks are that you want to improve.
SearchStorage.com: In closing, what's the single most important piece of advice you'd like to leave with IT shops considering solid-state storage?
Martin: You don't need a lot of solid-state storage to get a nice performance boost. For example, we've seen very significant performance improvements with only 5% or 10% of the total capacity allocated to SSDs. And we've seen 300%, 500% or more with a relatively small amount of SSD.