When using Progress SonicMQ, you can get different performance results on different platforms. Some of these differences can be explained with the disk setups used for the recovery logs. The documentation, mq_tuning.pdf, that comes with the installation has very good explanations for some of the general factors on pages 79 and 82, and I would recommend it as a reading if evaluating performance issues.
One of the issues, disk-caching, performed by the OS can be the factor you might be looking for. As the documentation indicates, the cache can increase performance but also might cause data-loss in case of failure. On a laptop windows XP system, the config screen might look as in the attachment. To evaluate your system, Progress provides a very useful stand-alone, pure-java testing code for disk performance simulating the real usage from their code-share site. I am attaching the code for quick review but the full package can be retreived from the Progress site.
The line "fd.sync();" is the actual place where the cache matters. You can run the test with this commetted out and see much different results especially if caching is enabled.
Although other factors, like disk block size, remote-disks, disk speed, queue save thresholds can be tuned, the disk-caching can be tested fairly fast and easily to see it affects.
No comments:
Post a Comment