[Linux-cluster] Performance of ES3+GFS6.0+GNBD+LOCK_GULM

[Kevin Anderson]

On Thu, 2006-01-26 at 15:44 -0600, Hong Zheng wrote:
>  
> 
> Hi, all
> 
>  
> 
> My configuration is as follows: two directors configured as LVS-NAT or
> LVS-DR, 4 GFS nodes as application servers which serve telnet and HTTP
> services, 1 GNBD node which has iSCSI connection to our shared storage
> server and 1 lock_gulm server. And before we ordered GFS we just used
> nfs mount to a file server which is currently a GNBD, but the
> performance is so slow for a certain application, since this
> application has multiple loops to access large number of data. We
> thought GFS is much more powerful than NFS, but not now. Do you have
> any idea to tune GFS performance or better configuration for such a
> situation?

Why are you using GNBD between the GFS nodes and the storage?  If your
storage is an iSCSI target, run iSCSI initiator on the GFS nodes and get
rid of the intermediate step.

Kevin