[rescue] Sun 711

[Chris Byrne]

> -----Original Message-----
> From: rescue-admin at sunhelp.org [mailto:rescue-admin at sunhelp.org]On
> Behalf Of George Adkins
>
> Is this unsound for figuring raw read performance?

Actually it can be depending on the composition of the file. What you really
want to do is take a 1 or 2 gig zip file full of JPG's or something similar.
That way you can be sure it's a dense file (no large areas of empty space or
long series of contiguous zeros). And that the transfer last long enough.
Anything liess than 10 seconds is insufficient, and personally I go for a
minute. Also you need to make sure that the file is large than the combined
system RAM, controller, and hard drive cahce if you want to see the true
sustained throughput to the spindles rather than cache throughput

I was also assuming write throughput for my calculations not read
theroughput. Read throughput can often reach over 90% of maximum if you are
reading from a (or a set of) non striped non parity mirror.

I hihgly recommend IOZone and Bonnie. THey will preovide a full profile of
boths reads and write, random and sequential, with and without cache flush,
for small to large file sizes and small to large record sizes (simulating
databse usage). You can then take those numbers and create a 3d I/O providel
in a graphing package and really see the I/O performance profile of your
system.

CHris Byrne