I created a new ft broker pair and made sure the disks are fast enougth.
There is considerable (10x) difference between not f(ault)t(olerant) and standalone ft broker.
I ran similar tests on windows platforms earlier but failed to reproduce the problem.
Any ideas?
===========================================================
Test case description.
Servers: 2 sun fire v490 sparc.
OS: Solaris.
Server names: sonic1, sonic2.
Test tool: testharness (all parameters take default values (except -b)).
Broker: Msg-3.
Messages: 1 kb. persistent and non-persistent.
Test description
Test 1. Msg-3 on sonic1 configured as plain messaging broker without backup.
Messages: persistent.
Test 2. Msg-3 on sonic1 configured as plain messaging broker without backup.
Messages: non-persistent.
Test 3. Msg-3 on sonic1 configured as fault-tolerant messaging broker with online backup broker on sonic2.
Messages: persistent.
Test 4. Msg-3 on sonic1 configured as fault-tolerant messaging broker with online backup broker on sonic2.
Messages: non-persistent.
Test 5. Msg-3 backup broker running on sonic2 in standalone mode.
Messages: persistent.
Test 6. Msg-3 primary broker running on sonic1 in standalone mode.
Messages: persistent.
Test No
persistent/non-persistent
Message throughput
msgs/sec
Cpu usage
%
Remarks
1.
ps
4940
~80
this not a problem of slow disks
2.
nps
4745-5015
~80
3.
ps
434
~10
4.
nps
4780
n/a
5.
ps
~500
~10
standalone brokers are 10 times slower than not fault-tolerant ones