On Mon, 2011-12-12 at 23:09 +0200, Uwe Geuder wrote: > On 12 December 2011 09:29, Jeff Hoogland <[email protected]> wrote: > > > Attached are the two outputs you requested, digging through them now to see > > if I can pinpoint the issue. > > > > Did you find out anything? > > I converted the outputs to csv, loaded them into an OpenOffice > spreadsheet, summed up each category of memory and compared your 2 > versions. The differences are really marginal, depending on the memory > category sometimes in favor of the old and sometimes in favor of the new > version. In terms of resident memory, which should be the most important > measure (no swapping has occured) the new version is even 792 KiB (~ 7 %) > smaller than the old one.
Thanks for looking at that; I was going to suggest something like this. As you've pointed out, RSS is the value that really matters. VSS doesn't matter at all. So any large (>25%) increases in RSS size between the dumps in any one library are interesting. But also that would indicate increased usage *in that library*, not necessarily in nm-applet. Now if you haven't changed any other packages/libraries on your system, but you've just changed nm-applet from 0.8.0 -> 0.8.1, then it may be that nm-applet is now using those libraries in a different way that results in a difference in memory usage. ie, it's actually not very straightforward to figure out this problem. Anyway, if we can figure out what might account for the change (if there is a large change) then we can look at what might be causing it. But if, as Uwe says, the RSS actually *decreases* in 0.8.1 then we've already won? :) Dan > Unless my conversion script really screwed up something and by accident > the bug just happened to level out your obvserved 110 MiB difference > such difference does not exist. > > If anybody wants my script and my spreadsheets to double check I can send > them by personal mail. I don't want the flood the mailing list with big > attachments, which are probably not of big interest for most > readers. (There are also other tools to read smaps files on the net, I > have never tried them.) > > Memory consumption in Linux is a tricky thing. There are many different > categories to measure (that's why smaps was added some time ago to show > them all or at least many of them). There is no single correct number. > If the tool you used to compute the 110 MiB delta shows only a single > number, are you sure the way the number is calculated has not changed > between your old and your new system? I assume you used the same tool in > the old and the new system, otherwise it's even more likely that you > ended up comparing apples and oranges. > > 110 MB difference looks huge by any measure. According to to my results > the mapped address space of the new version is "only" around 46 MiB. I don't > think any reasonable measure can be bigger than the mapped space. (The > old one is around 45 MiB, the difference 712 KiB) > > Regards, > > Uwe > _______________________________________________ > networkmanager-list mailing list > [email protected] > http://mail.gnome.org/mailman/listinfo/networkmanager-list _______________________________________________ networkmanager-list mailing list [email protected] http://mail.gnome.org/mailman/listinfo/networkmanager-list
