[Gluster-devel] About performance

Victor San Pedro vsanpedro at bioalma.com
Thu Jun 5 13:45:15 UTC 2008


I have been testing glusterfs with some old computers and I would like
to ask a couple of questions...
The PC's are 3 Compaq Desktop PIII 600MHz with 128 Mb RAM. They are
running Ubuntu without gdm, because of the low memory.
These machines are working as glusterfs servers.
The other two machines are P IV with plenty of RAM and processor. One of
then is working as glusterfs client and the other one as server.

The configuration is 1 Unify over 2 AFRs.

Configurations with AFR over STRIPE and opposite have not worked for me
in the machines previously specified. The debug does not reply an error,
but glusterfs is unabailable to work with these configurations.

With 1 Unify over 2 AFRs, I wrote some .jpgs on the glusterfs directory
ok, but when I tried to open one of them, it took arround one minute to
open it. This is from my point of view, a NOT very good performance in
deep. Which could be the problem???
I have been testing the memory % and the old cpus seem to be capable of
managing with the process load.
The debug reflect the read of the file, but there is like an intermedite
blank time. I do not understand why happens, and it delays the load of
the .jpg for too much time.
When these blank times happen, no process are working in memory neither
glusterfs process, and it is like glusterfs would have been waiting for
a process or something to finish in order to continue reading the file.
Do you know what is happening?
Could be a process attached to both files at the AFR in order to check
their synchronization?
Is it a normal behavior on glusterfs?
How could I improve my performance?
Fit these old machines that I have the specifications of glusterfs in
order to work properly?

Thanks for your help.

More information about the Gluster-devel mailing list