c0t0d0s0.org

News from the SLC flash front - Micron, Sun take NAND endurance to one million cycles:

Micron Technology Inc. (Boise, Idaho) has said it has worked with Sun Microsystems Inc. (Mountain View, Calif.) to develop a single-level cell (SLC) NAND flash memory technology that extends the lifespan of flash-based storage for enterprise applications and reaches a million write cycles.

Adam Leventhal wrote an really insightful piece about the potential effect of the Hybrid Storage Pool to the storage business in "Casting the shadow of the Hybrid Storage Pool":

What we have today though integrates flash in a way that changes the landscape of storage economics and delivers cost efficiencies that haven't been seen before. If the drives manufacturers don't already, it can't be long until they hear the death knell for 15K RPM drives loud and clear

When the performance of a storage architecture is ensured by SSD there isn´t really a need for low capacity 15k discs ... 7200 rpm (or even lower) drives with high capacity are more than enough.

Roger Bitar did some really interesting benchmarks with mysql on SSD. In his benchmarks he artificially constrained the availability of memory for the mysql buffers (aka cache):

We choose to test MySQL, the open source database (DB), using a simple MySQL benchmark called Sysbench. We populated the Sysbench table with 112 Million rows (around 26GB size) that fit on 1 SSD drive . We executed read-only queries while varying the buffer size. We mounted the file system in DIRECTIO mode to disable file system caching. We perfrormed the tests with regular HDDs and repeated them with SSDs.

There results were somewhat astounding. Of course a fully memory saturated database is the fastest way to implement a database (buffer cache at 24 GB). But with a database on SSD Roger was able to yield 96 percent of the performance of the in-memory database even with heavily constrained RAM space (buffer cache limited to 8 GB). The latency at memory constrained situation is vastly better with SSD than with rotating rust.

This leads me to an interesting conclusion: When you just compare the power consumption of SSD and hard disks (aka rotating rust), you forget a part of the equation. Let´s assume a FB DIMM takes 10 watts (it´s a little bit more, but 10 is easier for calculating). In the benchmark you were able to have almost the same performance with SSD (2 watts) at 8 GB as at 24 GB. Let´s assume 4 GB DIMMS. You would save 18 Watts (10*2-2) by using the SSD. Okay 18 watts isn´t a big deal, but now think about a 240 GB database.

You can design your database server in quite a different manner, when you take SSD into your considerations.

I´ve thought a little bit about the upcoming flash drives in our systems yesterday evening while summarizing a meeting with a customer regarding our Open Storage initiative. I really think that these system will change the business in storage.