> > Flash-based SSD's are not a solution, as random write speeds tend
> > to be as low as conventional disks, if not worse.
>
> That is not really the case for enterprise class SSD's.

Are you sure?

I understood that SSD's have to erase blocks much larger than a
traditional sector upon write. If there are large sequential writes
it works likes a charm, but random writes trigger much erase/write
overhead.

See:
http://64.233.183.104/search?
q=cache:YQ6b5qW1RxwJ:www.storagesearch.com/easyco-flashperformance-
art.pdf+easyco-flashperformance-art.pdf&hl=en&ct=clnk&cd=1

Although Jim has a point in that we are seeking solutions that scale
beyond a single machine, no matter how fast a single machine, a disk
with 0.1 ms write latency would keep up with a few thousand updates
per second, using traditional db approaches. Enough for most current
use cases, I think.