|
From: Eric S. <mas...@gm...> - 2011-06-26 10:06:43
|
Hi Nick You have understood perfectly what I have ment. Many thanks for your attendance of my question. Regards Eric Am 25.06.2011 23:51, schrieb Nicholas Nethercote: > On Tue, Jun 21, 2011 at 11:56 PM, Eric Schwarz > > When using 'massif' it is recommended for short programs to use > '--time-unit=B' option since o/w no or hardly any output is shown > using ms_print for visualisation [1]. My question is now regarding > the units of the x-axis. As far as I have understood represents > the unit the cumulated number of bytes allocated and freed when > option '--time-unit=B' is used. Snapshots are taken for > allocations and de-allocations, but also in-between of them. Since > the value of the y-axis (Each vertical bar represents a snapshot, > i.e. a measurement of the memory usage at a certain point in time. > [1]) and the value of the x-axis, which represents the cumulated > number of bytes allocated and freed, have a dependency to each > other, how is it possible that a parallel line to the x-axis is > drawn in the graph [1] since actually only a change in the heap > usage should generate a forward move on the x-axis. > > > I guess you're talking about this graph: > > 19.63^ ### > | # > | # :: > | # : ::: > | :::::::::# : : :: > | : # : : : :: > | : # : : : : ::: > | : # : : : : : :: > | ::::::::::: # : : : : : : ::: > | : : # : : : : : : : :: > | ::::: : # : : : : : : : : :: > | @@@: : : # : : : : : : : : : @ > | ::@ : : : # : : : : : : : : : @ > | :::: @ : : : # : : : : : : : : : @ > | ::: : @ : : : # : : : : : : : : : @ > | ::: : : @ : : : # : : : : : : : : : @ > | :::: : : : @ : : : # : : : : : : : : : @ > | ::: : : : : @ : : : # : : : : : : : : : @ > | :::: : : : : : @ : : : # : : : : : : : : : @ > | ::: : : : : : : @ : : : # : : : : : : : : : @ > 0 +----------------------------------------------------------------------->KB > ^^^^^^^^^ > And the horizontal lines, eg. where I've added "^^^" marks below the > x-axis? > > Those horizontal lines make sense with --time-unit=ms and > --time-unit=i. But you're right that they don't make sense with > --time-unit=B. Really, they should angle up or down. > > I'm not inclined to fix it thought; --time-unit=B is really only > there to facilitate regression testing, as it gives more deterministic > graphs than --time-unit=ms or --time-unit=i. > > I hope this answers your question! > > Nick > > Nick |