I use a lvs cluster for distributing http-reqs to 4 nodes who are connected via
an ocfs2-filesystem-cluster. The baseline of getting response of 4 very small
html files (only a few bytes long) to one server is about 9000 req./sec. Once I
increase the number of real servers the result isn't as high as expected - only
9.500 req./sec. The lvs hasn't any load; vmstat shows aroung 33000
interrupts/sec. during test whereas the baseline is about 1000.
CentOS 4.4 with kernel (2.6.9-42.0.2.ELsmp)
iptables _without_ connection tracking on
LVS-server: hp proliant dl360g4 with 2x 3 GHz
REAL-server: hp prolinat bl25p with 2x dc Opteron
Are there any kernel parameter that I can tune? Or other ideas?
Tanks at all!