SAS v SATA

[Edward Ned Harvey]

> Does anyone have an informed opinion as to the relative merits of SAS v
> SATA disk drives in a Linux box doing long sequential I/O for a small
> number of processes? It looks like the cost/GB is about a factor of 3
> more
> for SAS (both drives at 10,000RPM), and if the performance difference
> is
> only going to manifest itself in random I/O, or when many processes are
> competing for the drive, I can skip it for my current application.

I can say this:  I benchmark basically every opportunity I get, and I can't
measure a speed difference between SAS and SATA, except as explainable by
higher RPM's or drive capacity.  (With higher drive capacity, you get higher
bit density per track, and with constant rpm's, that translates to higher
data rate.  Also with higher drive capacity, you get lower fragmentation,
which again improves performance sometimes.)  

I can easily demonstrate SAS and near-line SAS to be more reliable, with
fewer failed drives over the life of the server.  IMHO, the reliability
factor is the advantage SAS has, and the reason it costs more.  Especially
for large sequential IO, you won't notice or care about the supposed speed
difference.

One more thing though.  There are no SAS SSD's available.  So if you'll ever
consider adding a SSD, use SATA.  I only know this because of desire to add
a SSD to my recently purchased SAS system.