Last posting we started looking into using Solid-State Disks (SSDs) in server systems. SSDs still cost in the ballpark of four to five times more than traditional spinning disks, but you can use them in lower-size applications such as system disks, e.g. to reduce reboot times. And the costs are coming down at about 50% per gigabyte per year, which is a faster decline than we are seeing in spinning disks.
Interestingly, boot time improvements, while dramatic, may not be the most impressive thing about SSDs. SSDs actually speed things up even more when loading applications. The reason is that a computer’s boot procedure involves a number of activities that are not I/O-bound, whereas application loads tend to be more limited by the speed of all the random-access activity that’s required.
I mentioned that SSDs are quite different animals and they have their own idiosyncracies. The biggest issue seems to be that they have a lower duty cycle when it comes to write operations, although this varies greatly both by product and by type. SLC-type drives (Single Level Cell) have longer lifetimes than the higher-capacity MLC-type drives (Multi-Level Cell). SLC drives tend to be used in enterprise applications, and the cheaper-but-twice-as-slow MLC drives tend to be used in the consumer space. Well, it’s not quite fair to say that MLC drives are twice as slow: their writes are twice as slow, but their reads are comparable to SLC drives. (Then there are SLC-MLC hybrids, just to make things more complicated!) Intel pegs its consumer-level drives (MLC) at a useful lifetime of 5 years, assuming 10GB of writes per day.
We’ll explore some other SSD quirks in the next post. Stay tuned.
