The results are still not flattering and are nothing close to native performance.  Unless GlusterFS has a "DRBD-like" option to delay writes over the network and to only read from the client side, I don't see how performance can ever improve much more.

After doing some client optimizations I added more to the score:

Start Benchmark Run: Sun Nov 29 00:37:44 PST 2009
 00:37:44 up 3 min,  1 user,  load average: 0.01, 0.02, 0.00

End Benchmark Run: Sun Nov 29 00:48:00 PST 2009
 00:48:00 up 14 min,  1 user,  load average: 14.43, 4.74, 1.77


                     INDEX VALUES
TEST                                        BASELINE     RESULT      INDEX

Dhrystone 2 using register variables        376783.7 20568131.5      545.9
Double-Precision Whetstone                      83.1     1157.8      139.3
Execl Throughput                               188.3      490.0       26.0
File Copy 1024 bufsize 2000 maxblocks         2672.0      563.0        2.1
File Copy 256 bufsize 500 maxblocks           1077.0      550.0        5.1
File Read 4096 bufsize 8000 maxblocks        15382.0    58805.0       38.2
Pipe-based Context Switching                 15448.6   543265.9      351.7
Pipe Throughput                             111814.6  1596349.1      142.8
Process Creation                               569.3     2885.6       50.7
System Call Overhead                        114433.5  2277096.2      199.0
                                                                 =========
     FINAL SCORE                                                      57.7
 

unixbench, glusterfs, openvz, quad, xeon, updated, optimized, configthe, flattering, native, quot, drbd, writes, improve, optimizations, benchmark, nov, pst, min, user, index, baseline, dhrystone, register, variables, precision, whetstone, execl, throughput, bufsize, maxblocks, context, switching, creation, overhead,

Latest Articles