Memory problems

[sgromoll]

Hi all-

My system has memory problems which i think are due to the nvidia driver. It seems that shortly after restarting X top reports that essentially all of my 1 Gb RAM is used, 280M of which is taken by X:

Mem: 1033164K total, 1005376K used, 27788K free, 291224K buffers
Swap: 979924K total, 3640K used, 976284K free, 521916K cached
...
PID USER PRI NI SIZE RSS SHARE STAT %CPU %MEM TIME COMMAND
17380 root 15 -10 280M 23M 4428 S < 4.3 2.3 3:03 XFree86

I'm using nvidia version 1.0-4191 with X 4.2.1 under Debian (running Gnome).

I'm not running anything too fancy (openGL games). This seems to happen even if i just run xmms, gnome mixer, and some xterms.

s it normal for X to hog ~300 mb, and is there any way to figure out if the nvidia driver is the culprit for my memory loss, and is it possible to fix...?

many thanks!

[bwkaz]

First, the main culprit is probably not X. It's probably the kernel caching filesystem data so that disk accesses are faster (and so that it doesn't take ~10 seconds to load up a directory with 8 files in it, like it does in Windows from time to time ).

To see the used/free counts without cached FS data included, you can take a look at the output of free. In the +/- buffers/cache line, in the "free" and "used" columns, will be the amount of physical memory in use by all processes.

Now, as for X, it's not actually using 280M of memory. That's what it says in the VSZ column, but VSZ is virtual memory in use, not physical memory in use. In fact, X is using 280M of memory, but >90% of that is swap.

The RSS column (resident set size) is a better approximation of the amount of physical memory in use. As you can see, X only has ~23M of its memory locked into physical RAM. And this is the normal behavior of X, at least on my machines too.

It is normal for Linux to use all the memory it can get its hands on. This is why the OS is much less tolerant of memory errors than certain other ones.

Oh, the cached data quickly gets discarded, too, if real processes need more memory. So don't worry about that.

[kickus]

I've experiensed the same phenomenon. I've thought it might be the twinview code which leaks memory, since it's maybe not that commonly used.
At least it's some part of the X or nvidia-module which does leak memory seariously.
It's not only because of caching, since the OS has started killing processes when it's really out of memory. When I restart X-server, I'll often see (in xosview for example) that about 800MB/1GB freed up.

I'm using kernel 2.4.20, nvidia driver: 1.0-4191. I've used kde 3.0.5 and 3.1.

Graphics card: GeForce4 Ti4200

I would really like a fix..

[bwkaz]

I haven't seen the kernel kill anything from being out of memory, ever. Even when I was using KDE 3 (it got to be too much eventually -- the eye candy was nice, but it took way too long to install it from source), I didn't see that.

But then, I'm not using twinview either. I'm using driver 3123 and kernel 2.4.20 with -preempt applied.