NFS "execution time" statistic. Anyone knows what it means?

Latest response

Hello all,

 

I'm looking into a performance issues for NFS volumes on RHEL 5.6; the volumes are served from a NetApp NAS device.  During times of heavy load on the NFS shares (approximately 6000 ops/sec) we are seeing IO wait times go up.  We are using NFSv3 with the following mount options...

 

rw,intr,nolock,tcp,nfsvers=3,rsize=32768,wsize=32768

 

When loads are light the nfsiostat command shows the "round trip time" and "average execution" time to be pretty much identical. For example...

 

op/s  rpc bklog

 642.25  0.00

read:   ops/s   kB/s    kB/op  retrans  avg RTT (ms)  avg exe (ms)
        0.500   2.125   4.250  0 (0.0%) 0.500         0.500
write:  ops/s   kB/s    kB/op  retrans  avg RTT (ms)  avg exe (ms)
        19.000  140.505 7.395  0 (0.0%) 0.579         0.618
When under load the "avg exe" value begins to be much longer than the RTT...
 
op/s  rpc bklog

2913.75  44.62

read:    ops/s    kB/s      kB/op  retrans  avg RTT (ms)  avg exe (ms)
         129.250  4136.312  32.002 0 (0.0%) 23.567        36.025
write:   ops/s    kB/s      kB/op  retrans  avg RTT (ms)  avg exe (ms)
         2211.750 70927.435 32.068 0 (0.0%) 4.050         65.749

 

 
 
Does anyone know what the actual difference is between RTT and Average-execution-time?
 
Thanks
 
Nathan
TL

Responses