From: Ballowe, Charles (CBallowe@usg.com)
Date: Tue Jul 02 2002 - 18:59:57 EDT
I was doing some tweaking and testing in my cluster (2xGS80) the
other day and noticed that relocating a filesystem to the node that
was actively using it raised the disk IO by almost double (25MB/s ->
40MB/s). I'm just curious what the internals are that produce that
level of overhead etc. How much overhead is there with CFS? Does
the member responsible for the disk do the physical reads and transfer
the data over the interconnect, or does it simply manage locks (the disk
is in an HSG80 controlled SAN) leaving the disk I/O to the the system
that needs the data? Are there any publicly available design docs on
CFS, or is it mostly considered trade secret?
thanks for any answers,
-charlie
Charles Ballowe /"\
Unix System Administrator \ / ASCII Ribbon Campaign
cballowe@usg.com X Against HTML Mail
x3896 / \
This archive was generated by hypermail 2.1.7 : Sat Apr 12 2008 - 10:48:45 EDT