[Linux-cluster] GFS poor performance

[Jonathan E Brassow]

It really depends on what you are doing.

GFS, in it's current state, favors large I/O.  You can get very good 
results if separate machines are simultaneously doing large block I/O.

For now, think of it like this: 1) the larger the block I/O, the better 
the performance 2) simultaneous non-contending I/O from different 
machines increase throughput.

You are probably testing from one machine with smaller I/O - violating 
both 1 & 2 above.  In which case, ext3 - a file system designed to be 
local (as opposed to clustered) - should be faster.

That being said, there are a number of improvements coming in GFS that 
will greatly improve workloads like the one you are testing.

Also, there are HA advantages to consider.  Performance vs High 
Availability...  Hopefully, in the not to distant future, you will not 
have to choose, but will get both from GFS.

  brassow

One way to think about the current state of GFS is to compare it to 
thread programming.  If you add threading to your program, but it has 
to do alot of locking and only runs on one processor; you've just 
killed all your performance.  If you add a processor (or machines in 
GFS's case), it is possible that you will see speedup.  If you increase 
your work chunks (increase I/O size) and reduces locking (reduce 
contention), you will see great improvements.

On Apr 27, 2005, at 8:53 AM, Patricio Bruna V. wrote:

> Its normal that GFS goes slower than ext3?
> i have the following scenario:
>
> 2 Dell 2.4G Xeon (dual)
> 2 HBA each
> 1 500GB Storage.
>
> i have configured GFS 6 and make some test on it and ext3 its far 
> faster.
> -- 
> Patricio Bruna
> pbruna at linuxcenterla.com
> RHCE/RHCI
> Jefe Soporte y Operaciones LinuxCenter S.A.
> Canada 239, 5to piso, Providencia, Chile
> http://www.linuxcenterla.com +56-2-2745000
> --
> Linux-cluster mailing list
> Linux-cluster at redhat.com
> http://www.redhat.com/mailman/listinfo/linux-cluster