On Fri, 2006-11-17 at 11:57 -0500, Alan D. Brunelle wrote: > Ming Zhang wrote: > > On Fri, 2006-11-17 at 11:28 -0500, Alan D. Brunelle wrote: > > > >> Ming Zhang wrote: > >> > >>> Hi Alan > >>> > >>> Thanks a lot for your quick reply. In fact I am reading your slide you > >>> made in 04/06. > >>> > >>> In your slide, page 20, there is a plot. For each activity, does this > >>> high mean some activities occurred while low means no activity? > >>> > >>> > >> Exactly - I've found that it is sometimes better to transform the data > >> presented in the .dat files (in particular, sometimes there are empty > >> sets, and it's great for post-processing xmgrace .agr files to have all > >> the sets with at least something in it). > >> > >> > > > > thanks. i think this plot is intuitive enough for me so far. especially > > in your example, capture that hot spot is quite easy once u read from a > > plot instead of from pure data line by line. > > > > > Good! > > i have another question. for example, each events are marked with a > > sequence number in blkparse output. but that number is not useful. i did > > not ready source code yet, but i believe there is internal connection > > for each event. for example, request X Q in event M, D in even N, C in > > event Z... So if output of blkparse can print out such info, then we > > have a complete chain to go from Q2C for each request individually. > > > This could be added - but in some instances there is a lot of data. Let > me take a quick look into it.
if this is the core data structure http://lxr.linux.no/source/include/linux/blktrace_api.h?v=2.6.18#L76, will a u32 parent_sequence will enable this feature? > > > > also a usage issue. > > > > # btt -i x.blktrace.0 > > > > can show me a lot of info but blkparse show me nothing on attached > > trace? > > > > # blkparse -i x.blktrace.0 > > > > Throughput (R/W): 0KiB/s / 0KiB/s > > Events (x.blktrace.0): 0 entries > > Skips: 0 forward (0 - nan%) > > > > i just check out the git tree. > > > you don't run btt on the raw blktrace data - you use blkparse -d <file> > to create a binary stream that btt can read. why? i ran btt against it and i can generate a lot of data. my problem here is that blkparse could not find any data in that raw trace. > > > > ps, u prefer to send private email to u or always cc to list? I always > > ask my list subscriber to cc lists. > > > Ah, switched e-mailers, I should cc the list... > > Thanks! > > > > Ming > > > > > >> Alan > >> > >>> Ming > >>> > >>> > >>> On Fri, 2006-11-17 at 10:53 -0500, Alan D. Brunelle wrote: > >>> > >>> > >>>> Ming Zhang wrote: > >>>> > >>>> > >>>>> Hi All > >>>>> > >>>>> I read from a presentation that people can use xmgrace to explain the > >>>>> btt results. Could anyone shed some lights on this topic? Thanks! > >>>>> > >>>>> Ming > >>>>> > >>>>> > >>>>> > >>>> One of the outputs from btt is a file called <file>.dat - which contains > >>>> sets describing Q and C activity. [Basically: when IOs enter the block > >>>> IO layer, and when they are completed.] If you specify the -q or -l > >>>> options, you can get individual IO information in appropriate grace data > >>>> file format for the total time each command takes (q2c) or the time > >>>> spent in the driver and on the device (d2c). > >>>> > >>>> You can import these files directly into xmgrace. > >>>> > >>>> Alan > >>>> > >>>> > >>> > >>> > - To unsubscribe from this list: send the line "unsubscribe linux-btrace" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
