All, 
 
We use log4j across a number of services where many servers write to a shared 
disk. The net impact of the log writes contributes to the total number of 
physical disk writes that the disk can support. This has become our overall 
system bottleneck (NOT THE CPU)
 
Has anyone tried tuning the logging system to the behavior of the underlying 
disk layout?
 
__Rough Algorithm Suggestion__
Create a buffer that is aligned with the optimal write size of the physical 
disk medium. You can find this out for your system via IOmeter; 
http://sourceforge.net/projects/iometer.
 
Append the random sized log messages into the aligned buffer and only flush to 
disk the buffer fills.
 
This would reduce the number of physical disk writes to the underlying medium 
and thus improve the overall system performance.
 
Thanks
dg
 
 

Reply via email to