The write cache capabilities of the XcelaSAN are an important and unique capability. Most caches, because of fear of reliability, are only read based which means you are only accelerating a portion of of your workload. While in many environments there are more reads than writes those writes are the slowest function that a storage system has to perform. Caching those reads has several benefits. First and foremost the performance boost is immediate, there is no need to “warm” the cache up. The moment you execute a storage write it is cached, mirrored to the second XcelaSAN appliance and the application is told that the write has been committed.

Compare these to some of the auto-tiering or read only caching technologies that are becoming commonplace in storage systems today. They have to wait and analyze data to make sure to figure out what data is needed in the higher speed tier. Write caching if done with high availability brings an immediate performance boost. Another important differentiator with the XcelaSAN is that it is DRAM based and uses flash SSD for backup of the DRAM area. DRAM unlike FLASH has no performance difference when having to deal with heavy write traffic. It also makes a high cache turn over a non-issue since there is no concern about wear leveling or garbage collection.

There is a whole string of benefits from write caching that we could get into but are beyond the scope of this note. For example the unit can do write avoidance which is when they can hold write activity on a very busy file in cache for as long as possible. In this situation it is not uncommon for the same block to get updated over and over again. They only write that block either as the cache fills or as the block cools, meaning a dramatic decrease in the amount of writes that go to the actual mechanical drive array.

Of course the XcelaSAN also does all the read caching you would expect, but on block data which is not as commonplace as some of the NFS accelerators that we have discussed in the past. The unit lets you decide how the cache will be used as a percentage of read and write and you can determine which LUNs are cached by the unit. This allows you to make sure that your general purpose file server doesn’t steal valuable cache memory from your database application environment.

Thanks to its high availability capability, the XcelaSAN integrates seamlessly into your existing infrastructure. When inserting into a highly available SAN simply bring down one side of your SAN and insert one appliance. Bring the first side back up, then bring down the second side, insert the H/A XcelaSAN pair, and bring up the second side. Dataram claims an hour of implementation time and then again thanks to write caching, instant performance boost.

George Crump, Senior Analyst

ExecEvent Briefing Note

The Dataram XcelaSAN is now shipping. This is as far as I know the first commercially available block level solid state caching appliance. There are some storage virtualization systems that do this but you have to switch to their way of doing things. The XcelaSAN simply makes your current storage system go faster (by as much as 24X) and you don’t have to learn new tools or data protection strategies. By working on block storage it delivers a performance boost across a broad range of applications in your environment. To me the part that stood out was that the solution is highly available, making it safe for not only read operations but write operations as well.

Dataram is not a client of Storage Switzerland

Storage Swiss Take

We have been following the progress of Dataram’s XcelaSAN for almost two years now. They took a very cautious approach to producing a product and waited until it was right before releasing to the market. The result is a product and a company that I believe you can trust. There are amazing benefits from leveraging solid state disk for caching, even more when you trust the system to do write caching. That trust has to come with some high availability capability which is difficult to get just right. It seems that Dataram has cracked that code.

Storage Switzerland was at the ExecEvent and as with other shows we did our best to bring live briefing updates. As is always the case and even more so in the rapid fire format, consider these briefing notes as raw feeds with the goal being to get the information out quickly.