For most applications, the storage industry is fairly adept at delivering requisite performance. All, that is, except for large data set processing. Think: Financial market modeling, or digital image rendering, or seismic analysis for the gas and oil industry. For these applications, thousands of servers churn away for days before the job is finished. And when there's a lot of data fetching, the speed of the storage system is critical, and in most cases, currently inadequate. Gear6 thinks it can help. The company has been around since 2002 but has been focused on the problem of storage access speed since 2005. Announced in June, but shipping since January, its cacheFX products sit in front of NFS filers (CIFS and other protocols are on the road map)with lots of networking bandwidth and lots of cache memory. Lots.
The company's smaller cache appliance packs 250 Gbytes, and its larger weighs in with 500 Gbytes (apparently two of the smaller ones stuck together). It claims a performance for its appliances of 250,000 IOPS and 500,000 IOPS respectively, each with a response time of .5 milliseconds or less. Half a Terabyte not big enough for you? Don't worry, the appliances are meant to be stacked together to build larger systems. The cache coherency software is the company's secret sauce.
The system is a write through cache, so no changed data is held in the system for longer than it takes to transmit the data to the filer. That means read-mostly applications are the ones that will primarily benefit from this technology.
While clearly not useful for many applications, a technology like this makes immediate sense to those who need it. It takes environments with hundreds or thousands of servers before the advantages will make sense. At its rated latency, the system couldn't possibly meet its peak IOP rate without at least 125 servers beating on it. In practice, it should require a lot more.
The cacheFX is a beefy product with complex software, but a rather simple and direct use. For those who need it, the $400,000 entry price won't be a deterrent. If you can cut a financial simulation's run time in half or better, the TCO upside won't be hard to figure.