Storage Tiering with Flash and DRAM

-written by Tom Mayberry, Dataram Storage Blog Team Member

When considering the storage hierarchy, it is important to understand why it makes sense to buffer the Flash tier with DRAM given the price, performance and properties of the various solid state technologies available today. The primary solid state elements in the storage hierarchy are DRAM and Flash.

DRAM is faster than Flash (by up to 100x), and doesn’t “wear out” like Flash does. There is a finite amount of read/write cycles associated with NAND Flash, and that doesn’t exist with DRAM. Flash has struggled with data integrity (MLC vs. SLC), which contributes to its slower adoption. But Flash is non-volatile (a big plus), vs. DRAM which loses all data when power is removed.

In addition to its non-volatility, the pluses of Flash are that it costs less and you can configure larger storage capacities. An Intel 300GB Flash SSD costs about $550 (<$2.00/GB) while DRAM cost about $9.00/GB (8GB RDIMM), about 5X cheaper. It is a classic trade off of speed vs. capacity in the storage hierarchy.

In the hierarchy, Flash comes in above of HDD, but below DRAM. And DRAM is below SRAM cache (L3/L2), or CPU cache (L1/L2). Therefore, it makes perfect sense to have DRAM buffer or cache Flash, and Flash to provide the first tier of non-volatile storage, with mechanical HDDs the final non-volatile tier. So even in an array with a large compliment of Flash drives, it still makes sense from a performance, wear, and price perspective to buffer the Flash tier with DRAM.

This entry was posted in Storage Posts. Bookmark the permalink.

Comments are closed.