[GRLUG] Solid State Drives

[Michael Mol]

On Tue, Dec 2, 2008 at 11:02 AM, Collin Kidder <adderd at kkmfg.com> wrote:
> David Pembrook wrote:
> Actually that is the beauty of it. A compressed swap would transfer some
> of the bandwidth cost to the CPU instead of the hard drive. If, say, the
> harddrive can maintain 60MB/s and the CPU can compress to half size then
> you effectively get 120MB/s throughput. In a fast multicore machine
> (which is common nowdays) the CPU usage is a pretty small concern.

Keep in mind that most compression algorithms are stateful and
stream-based, and very difficult to multi-thread.  If you treat each
memory page as a separate stream, and keep a processing thread pool
equal to your number of available cores, you'll be able to take
advantage of your shiny 128-core chip, but your actual compression
will be very poor; Compression tends to improve as you increase your
volume of compressed data in a stream.

You could tweak the rest of the system to improve performance, though.
 Try zeroing out pages of RAM between uses.  Lots of zeros make for
highly compressible data. :-)

-- 
:wq