Why you should avoid placing SSDs in traditional Arrays
Sun blogger Anatol Studler makes that case in a recent blog that you should avoid placing SSDs in traditional arrays.
A traditional array has a controller with a certain fixed amount of compute power and I/O capability. Placing fast SSDs into an array designed for traditional disks is, “is quite surprising to me, as it is comparable to place a 8-cylinder bi-turbo engine with 450HP into an entry level car.”
Anatol continues, “Traditional midrange arrays are developed to handle hundreds of traditional (15k RPM) harddisk drives. A traditional harddisk is capable of running about 250 IO/s. Now if we compare this with the actual enterprise class Solid State Disks available on the market, a single solid state disk can do about 50k IO/s read or 12k IO/s write. So in fact it is about 100x faster than a 15k RPM harddisk.”
A fast array controller capable of delivering 500k I/Os would hard pressed to keep up with just 10 SSDs.
In addition, SSDs have a very low latency. A trip through the FC network, cables, switches and array controller for a single I/O will consume as much time as it will take the SSD to deliver tens of thousands of I/Os.
Anatol poses the the question, “So, where do I place the SSD technology?” and answers, “As close as you can to the server!”
His suggestion is to directly connected SSDs to the server or place them in SAS connected JBODs such an the J4000 family introduced by Sun last month.
With ZFS, you can create a hybrid storage pool. Use write biased flash for the ZFS ZIL (ZFS intent log). Use reach biased flash for large capacity, persistent cache (ZFS L2ARC); Use JBODs for capacity/density. Later, use HSM and tape arrays for capacity without power consumption.
More Information
ZFS Second Level ARC - L2ARC - Testing Shows 8x More Throughput
Can Flash Memory Become the Foundation for a New Tier in the Storage Hierarchy?
Automatic Data Migration (ADM) with ZFS — OpenSolaris HSM for Solaris File Systems, Including ZFS