On Wed, Jan 09, 2013 at 04:05:26PM +0100, Paul Bijnens wrote:
I'm still not really out of the problem either:
On 2013-01-09 14:21, fred smith wrote:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND11159 fredex 16 0 263m 149m 10m S 0.3 3.8 1:36.87 clock-applet
263m VIRT: We understand that virtual memory is sum of all, including the read only/executable parts of lots of shared libraries which are not using many resources in fact because they are backed by the real files.
149m RES: The parts that are really in use by "this" program, but including the parts shared by many others of the Gnome package that are running. (I put "this" between quotes, because someone seemed to believe that top or ps or some other program adds memory sizes in in "unfair" way to some processes only, while they should have been distributed over many. Actually I can't find any reason, or explanation of that.)
10m SHR: of those 149m that are in use, 10m is the shared portion.
That leaves 139m allocated exclusively by the clock-applet. Incredibly large yes.
What is that clock-applet doing with 139m?
I find that to be a really good question. When I first posted, the system had been up for (from my memory) something like 29 days, and the RES column for clock_applet was some number well above a gigabyte. I can imagine that those who say it's just being reported wrong may be right, but if so I don't see why that number would continue to grow. in fact it's also a fact that the amount of swap in use had grown above two gigs, which is definitely NOT the norm on this system. So I still think it's being leaked, _somewhere_.
Or is that just because you do have lots of RAM, and in fact, lots of those pages could be swapped out (= for executables like shared libraries, just be removed; they can be swapped in from the file itself), but never had to be swapped out/deleted, because the OS still had unused pages.
I have rebooted my workstation (due to postponed kernel upgrade) yesterday. I have now:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND 5908 paulb 15 0 283m 13m 9496 S 0.0 0.3 0:25.29 wnck-applet 5950 paulb 15 0 268m 10m 8968 S 0.0 0.3 0:54.00 clock-applet
and I can't find any process that is taking large amounts of memory. Firefox & Thunderbird are using considerable amounts but not more than usual. I did not run any real large program yet (swap use is 0 now), and did not do heavy memory using programs except ff & tb. So why is my RES on 10m now? Probably because the clock-applet did not even touch more of the shared libraries (totaling 268m possibly being touched).
As I said, the problem of the large clock-applet magically disappeared for me somewhere between 5.1 and 5.5 or so. I remember fiddling with the seconds indicator, changing every option to default, monitoring the mem over several weeks, seeing it increasing, but seeing no trend/reason/cause. Gave up on searching intensly. And then some time later noticed the problem disappeared. Yesterday I thought wcnk-applet had the enormous amount of RAM, but I think I was looking at the VIRT size only then. VIRT is a useless indicator here.
Inspecting /proc/PID/smaps of such a large process may reveal something?
well, there's a LOT of stuff dumped when one cats the file. but I have no adequate expertise to figure out what it all means.