NetApp’s August 21 Flash Accel read-only flash cache announcement is good as far as it goes, but it is a tactical product that misses strategic direction, writes Wikibon CTO David Floyer in his latest Professional Alert, “NetApp adds Server Flash Cache to its Arsenal – but misses Strategic Flash Direction.” The Alert provides a full analysis of the details of the announcement.

The good news, Floyer writes, is that NetApp has chosen flash storage server card vendor Fusion-io as its Tier 1 vendor for its new flash cache. The bad news is that the cache is read only, with all data written through to NetApp disk storage before moving to the cache. Any attempt to write to data in the cache locks that data, forcing applications to wait until the changes are written to disk and then fed back to the cache on the server. This slows the entire system back down to the speed of spinning rust, which is an order of magnitude slower than flash.

The issue here is that there is no need for this restriction. Fusion-io’s cards are fully read/write on the server and are in fact designed to be used as flash-only storage, with no connectivity to a background disk storage box. The advantage is blazing speed, literally orders of magnitude faster than the highest performance any disk system can deliver, for both reads and writes. The disadvantages are that flash costs more on a cost-per-Gbyte basis and that the storage is most efficient when working with an application on the server where the data is stored. For Tier 0/Tier 1 data such as active transactions that have very high access rates and where cost-per-IO and absolute read/write time are the important financial measurements, flash is much less expensive than disk.

By artificially limiting Flash Accel to a read-only configuration, it has limited practical use to data that is either read-only or has very few writes but has a large volume of reads. This is at best a small subset of Tier 1 data. In this it resembles EMC’s initial flash cache offering announced last spring, which was also read-only. Since then however EMC has purchased XtremeIO and IBM has bought Texas Memory, moves that portend announcements of flash strategies that should include integration of read/write flash cache on servers as part of multi-tiered storage systems.

So far NetApp has not provided any indication of movement toward a similar strategy. Therefore, Floyer writes, “Users should be wary of choosing NetApp as a long-term flash strategic partner while it continues a cache-only strategy.”