ext3 documents system performance tuning: what metrics to check
I have a high-IO application that is functioning, and also scaling well to day. Over the previous couple of months we've been attempting to overlook the roadway and also forecast where our next traffic jams will certainly take place. Among them is undoubtedly the documents system.
We are presently monitoring
- Room readily available
- Read procedures per 2nd
- Write procedures per 2nd
This appears a little bit also thin to me. What else should I be seeing? I am not also certain what the 'yellow line' would certainly be for the operations/second.
- Look at the read line up dimension, if your application is very arbitrary, after that fine-tune the readahead in
/sys/block/<dev>/queue/read_ahead_kbto guarantee you're reviewing information you require , not information the OS assumes you require.
- Switch over to the target date scheduler if you have not currently
- make use of the
noatimeplace alternative unless you're organizing a mail spindle
- place with
data=writebackif you've obtained excellent back-ups
- watch on your directory site dimensions, certain hashed directory site inodes aid, yet if you can hash the information on your own after that you'll get even more regular outcomes