You are here
Apples & Oranges: Caching in the Server v. Caching in Storage
Since we launched our company a couple of weeks ago, we still enjoy seeing FlashSoft mentioned in an article, even if it's just a mention like "other companies with flash-based caching solutions include FlashSoft…" After all, we'd been working under the radar since 2009, and only introduced ourselves once we had a shipping product, some customer successes, and partnerships with industry-leading vendors. So even a brief mention in a blog or publication is a nice reminder that we're now in the public eye.
But there is one thing that tweaks us a little, and probably every new technology developer has had this feeling. It's the sense that we've been put into a category with solutions that are not much like ours. Of course, the job of the industry press and analyst community is to help busy people understand emerging categories. So we sometimes find ourselves in a catch-all category like "flash-based caching for enterprise storage."
And that's where we think there's an “apples vs. oranges” difference that we should point out. If all you need is a general idea of who’s working on enterprise flash for caching/tiering, maybe the difference is academic. But if you are an IT professional investigating how best to use enterprise flash to improve performance and scalability, we think the difference is absolutely critical.
What is this critical distinction? Very simply, FlashSoft uses SSD as a cache in the server. What this means is that the SSD is attached to the sever via PCIe or SAS, or even SATA. That's it: our software, installed in the server, enabling your SSD as a persistent read-write cache for hot data. In the server. That's the apple.
What's the orange? Anything using SSD as a cache that is not directly attached to the server. Whether it's a flash-based tier in a storage array from a major vendor or a flash-based SAN appliance from a startup, an SSD cache inside storage hardware does not make it "the same kind of thing" as server-attached SSD-based cache.
Why does the difference matter? Well, performance for one thing. A PCIe card in the server delivers data at speeds from 10 to 20 microseconds. Traversing a SAN takes ten times that long in the most optimal configuration, and in the real world optimal configurations are rare. (Let’s not even talk about data writes – that’s a whole blog post in itself!)
Another reason why the difference matters is ease of adoption and implementation. As an example, here’s what it takes to evaluate FlashSoft SE in a server: (1) Identify the server where IO is bottlenecked, (2) install your SSD (or SSDs in a RAID), (3) install FlashSoft and identify the SSD that will be used for caching and the volumes to be accelerated. (4) Done.
We also see an “apples and oranges” difference in how people define the IO bottleneck problem. Many people have become used to thinking that the question we should ask is “how can I make my storage IO faster?” But the question we should be asking is “how can I make sure that each of the key limitations on my servers’ performance and scalability – CPU, memory and IO – are proportionately balanced?” When we ask ourselves this more fundamental question, we realize that our mission shouldn’t be to relieve the pain of the IO bottleneck from the storage tier upward, but to eliminate the IO bottleneck from the compute tier in the first place.
If you’re seriously investigating caching solutions that use enterprise flash, this distinction between server-tier solutions and storage-tier approaches is one starting point to consider as you investigate your options. If you’re betting your organization’s resources and your personal success on the outcome of your investigation, the question is hardly academic.
- - - -
End note: not every overview of the market glosses over the differences among solutions. Mark Peters at ESG has written an overview that's very broad in scope, and also provides sufficient detail to understand what makes each player unique. It's only available to ESG subscribers, but it shows that the industry landscape can be described without losing sight of important points of differentiation.
