[Swap] I swap too much

Question

Hi,
I have been experimenting swap issues for about one month. I swap for no reason, while I still have free RAM memory, and while the parameter vm.swappiness = 1 in my /etc/sysctl.conf.
Have a look:
free -h
total used free shared buffers cached
Mem: 3,8G 3,5G 272M 0B 106M 2,2G
-/+ buffers/cache: 1,2G 2,5G
Swap: 2,7G 93M 2,6G
Do you have an idea about how to solve that ? I had not some months ago.
Sincerely
Last edited by djipey (2013-06-23 11:04:04)

Answer 1

Still, swapping is usually an indication that you need more RAM, as it is running out.  So in your case in which it is obvious that you are not running out of RAM, I think you should do a "swapoff -a" just to see if your system starts to hang or starts OOM killing things even when the RAM is not full.
[offtopic]
I don't necessarily think that there is a "correct" way between the kernel command line and sysctl.{conf,d}.  I just try to configure things with the native tools of my system, since they are going to be running anyway.  I think for me, it just serves as a way of learning about those tools.  Using the kernel command line will actually produce the same end result, I just like to keep in under control.  As an example, it is possible to actually fully disable udev and specify every module to be loaded on the kernel command line, but that just seems silly since there are other ways to go about it.  Udev has become pretty central to our systems in terms of harware initialization, and it works pretty well, so why not use it?  And even if you have a reason not to use it, it would just make more sense to create a list of necessary modules and put them in /etc/modules-load.d wouldn't it?  Thats just my two cents.
If you are using a script to change power parameters between AC and battery power, then none of those parameters are going to do anything anyway since the script is just going to handle things itself anyway.  I used to use a script to change parameters on my system, but honestly, with the power of today's machines, I really see no reason to enforce inefficient policies when using AC.  I think of it in terms of the usability of the hardware between the two options.  Can you really tell a difference between the perfrmance of the computer when you have it plugged in vs. not plugged in?  I couldn't, and I would imagine that most people won't be able to apart from the possibility of benchmarks and certain special cases (like real-time audio uses).
So what I have done is I just used powertop (and various things I found about the interwebs) to configure my laptop to be more power friendly.  They are set to those parameters by a combination of tmpfiles.d, udev/rules.d, and sysctl.d, and they do not change.  Alternatively, you could just redirect values into the /sys and /proc files, and these parameters can be found by using powertop's ability to generate an html report (see the powertop man page).  But again, by doing it with the tools that were on my system, I was able to learn how to use sysctl, tmpfiles, and how to write basic udev rules.
I believe I have told taylorchu before about the incorrect rc6 setting in his 3x battery life thread (powerdown is taylorchu's script, right?), but apparently it was never changed.  Just take a look at the modinfo of i915.  rc6 has a 1 value, rc6p has a 2 value, and rc6pp has a value of 4.  So if you want all three, you add those up and get 7.  It is not imperative that you make this change, but if it were me and I was trying to implement power saving, I would want the potential for the greatest level of power saving that I could.  I think that it is just a common misconception that rc6 should be a 1, because that is what ubuntu does.  Though I feel like this is just a rememant of when rc6 was in its infancy and both didn't have the deeper sleep states and didn't necessarily turn it on by default on all hardware.
[/offtopic]