[Linux-cluster] gfs perfomance tuning

[Mital Patel]

Hi,

I am looking for suggestions on how to increase the speed of our GFS
configuration.  We are using GFS on a cluster of 4 web servers with a
EMC AX150i iscsi server backend.  Cluster/GFS is setup and working
properly, however it seems to have horrible performance when multiple
nodes attempt to create files in the same shared directory.  Our
application require us to create many small files on the shared drive,
which according to this site,
http://kbase.redhat.com/faq/FAQ_78_3152.shtm is not the best environment
to use GFS.  I am exploring alternatives, but if we can get GFS to a
point where it's usable, we want to stick with it and hope GFS2
increases the performance even more.

I have read through some previous threads and it seems like I can eek
some performance by increasing the gfs_scand interval, is that correct?
Are there any other settings that would help with performance?

System Info:
Centos 4.5
kernel 2.6.9-55.0.6.ELsmp
GFS-kernel-smp-2.6.9-72.2.0.7
GFS-6.1.14-0
dlm-1.0.3-1
dlm-kernel-smp-2.6.9-46.16.0.8


gfs_tool gettune:
ilimit1 = 100
ilimit1_tries = 3
ilimit1_min = 1
ilimit2 = 500
ilimit2_tries = 10
ilimit2_min = 3
demote_secs = 300
incore_log_blocks = 1024
jindex_refresh_secs = 60
depend_secs = 60
scand_secs = 5
recoverd_secs = 60
logd_secs = 1
quotad_secs = 5
inoded_secs = 15
glock_purge = 0
quota_simul_sync = 64
quota_warn_period = 10
atime_quantum = 3600
quota_quantum = 60
quota_scale = 1.0000   (1, 1)
quota_enforce = 0
quota_account = 0
new_files_jdata = 0
new_files_directio = 0
max_atomic_write = 4194304
max_readahead = 262144
lockdump_size = 131072
stall_secs = 600
complain_secs = 10
reclaim_limit = 5000
entries_per_readdir = 32
prefetch_secs = 10
statfs_slots = 64
max_mhc = 10000
greedy_default = 100
greedy_quantum = 25
greedy_max = 250
rgrp_try_threshold = 100
statfs_fast = 0





Mital