Hi,

we have a write performance problem that I hope maybe someone could shed some 
light on.

we have two linux servers, roughly equivalent regarding CPUs and memory 
running the same OS version (Kernal 2.4 SuSe 8.1) and the same version of 
sapdb with almost identical configuration (as far as I can tell relevant for 
write performance). We noticed a big difference in write performance (we have 
a little java tool that issues many inserts and commits each 100 records), 
i.e. one system is 5 times faster than the others, while read performance is 
nearly identical. The main diffference is that the fast system uses a RAID5 
system while the slow system uses a fast scsi disc. since this is an obvious 
thing to look for we performed filesystem benchmarks using the IOzone 
benchmark tool (http://www.iozone.org) and found a big difference in the 
write (in IOzone terms, writing to a newly created file) performance for 
small block sizes, that would explain the difference in database write 
performance as the order of magnitude of the difference is roughly that of 
the difference in insert-performance.

However, the rewrite performance (in IOzone terms, writing to an already 
existing file) is even better for the "slow" system. I thought sapdb writes 
to existing files and therefore that number is the relevant one destroying my 
theory that disc write performance is the problem.

Another strange thing is that on the "slow" machine stopping the database also 
takes ages compared to the "fast" system.

Can anyone explain that or take a guess?

Thanks,

Robert



 
_______________________________________________
sapdb.general mailing list
[EMAIL PROTECTED]
http://listserv.sap.com/mailman/listinfo/sapdb.general

Reply via email to