Bug 761817 (GLUSTER-85)

Fri Dec 12 12:23:54 2008  	 guru - Ticket created

I am attaching the result of an experiment I did based on Avati's
request. I have tabulated the times taken for the three commands:

cp -a /usr .
du -hs usr
find . -type f | xargs cat > /dev/null

The server had posix and posix-locks.
The client had one afr over two servers (in case of AFR) or just a
protocol client (in case of point-to-point). The client in all the cases
was being run on a different machine than the servers.

The backends were cleaned each time.

Note that I did the exercise for IB and TCP.

All timings in Seconds
cp du find | cat
AFR TCP 892.16 92.72 721.48
AFR IB 604.56 96.62 231.47
P2P TCP 399.04 33.77 400.79
P2P IB 337.79 26.14 129.65

--------------------------------------------------------------------------------
#   	Tue Apr 07 18:17:17 2009 	gowda - Correspondence added 

avati,
why was this test recommended? what was the conclusion of the test?

did any of the enhancement, bug-fix, performance optimization that went
into glusterfs depend on this test? if yes, please update the ticket
with the details (or overview) of the changes that went into glusterfs
as a result of this test.

-- 
gowda