From: Maxim Uvarov <muvarov@xxxxxxxxx> Subject: Re: [RFC PATCH 0/4] makedumpfile: cyclic processing to keep memory consumption. Date: Mon, 4 Jun 2012 12:29:07 +0400 > 2012/6/4 Atsushi Kumagai <kumagai-atsushi at mxc.nes.nec.co.jp> > >> Hello Maxim, >> >> On Thu, 31 May 2012 10:50:39 +0400 >> Maxim Uvarov <muvarov at gmail.com> wrote: >> >> >> After received your mail, I measured executing time with the patch below. >> The result below was measured in 5GB memory machine. >> >> Result: >> a. makedumpfile -Kcd31 >> excluding time: 6.55 [sec] >> writing time: 5.89 [sec] >> >> b. makedumpfile -cd31 >> excluding time: 0.21 [sec] >> writing time: 5.82 [sec] >> >> I don't think this result is good, the prototype has some points to be >> improved for performance. >> (e.g. buffer size, the way to exclude free pages, etc...) >> >> However, I think getting a bit of speed down can't be helped to keep >> memory consumption. >> >> > I'm asking because I have machines with several TBs of ram. And filtering > and saving data from vmcore takes long time. (more then 30 minutes). I > wonder if somebody already work on performance. Probably we can do SMP > optimizations. > > Maxim. Hello Maxim, I'm now benchmarking for three kinds of pralell compressions: no compression, lzo and snappy. I'll post its result after completing it, by Friday? if possible. Also I'll soon post a patch for filtering free pages in constant time based on mem_map, just a small extension of the patch Kumagai-san posted previously, by which filtering score for -Kcd31 above would be improved so much; it now repeats tracing free_list (memory size / cyclic_size) times. Thanks. HATAYAMA, Daisuke
