From: <yam...@va...> - 2010-04-13 07:17:46
|
hi, > Hi, Takashi-san > > 2010/4/13 YAMAMOTO Takashi <yam...@va...> > >> hi, >> >> >>> > And I like to issue the problem about ioband >> >>> > Currently, ioband controller has big overhead in controlling I/O >> >>> bandwidth. >> >>> > for example, >> >>> > the overhead is seriously arised according to the increasement of >> >>> processes >> >>> > or process grorups(using cgroup) >> >>> > especially, weight policy is more serious now. >> >>> > If you analyze the source code these days, can you take a >> consideration >> >>> for >> >>> > that? ^^ >> >>> >> >>> i have no idea right now. i'll keep it in mind. >> >>> do you have any numbers of the "big overhead"? >> >>> >> >> >> >> Yes, I attached evaluation result file(excel file) >> >> So, you can refer the file to understand the situation. >> > >> > thanks. >> >> can you explain a bit? >> >> what is X-axis? >> >> X-axis means time as showed in graph. > the result was picked out in every second > > >> what do total-BW, range-bw, weight mean? >> > > TotalBW means origianl disk I/O bandwidth without ioband controller. > and range-bw and weight are the total bandwidth by control policy that I > applied in the test. do you mean that you run the same workload 3 times with different configurations? (no ioband, range-bw, and weight) do you see the similar performance drop once you configure ioband, regardless of values of min_bw, max_bw, etc? otherwise, can you show me those values which you used for this graphs? YAMAMOTO Takashi |