This is probably very old news for many, but I just found a nifty little program for the first time: cpulimit.

I convert a lot of video files. A favourite source supplies only .FLV, but some family  members are Apple-bound and require .MOV, so I convert the videos using avconv and store the results on our fileserver.

It had annoyed me for some time that while avconv was running on my system, I was effectively prevented from doing anything else. The window manager became jerky, froze for several seconds at a time, video playback was unusably jerky, sometimes stopping for many seconds at a time, or the image would lose sync with the sound. Even the mouse sometimes stopped being responsive for a second or two.

Using top, I saw that avconv was using 150% of the CPU (I have a dual-core computer), sometimes more. I tried using renice to control avconv‘s excesses, but even with the avconv process set to a nice value of 19, the high CPU usage and symptoms continued.

Finally it all got too much and I went looking for a solution – and found cpulimit.

This delightfully simple little program lets you specify a process by name or ID, and what percentage of total processor power that process is allowed to use.  I use it now on avconv, and avconv now runs a bit more slowly, while everything else (including me) keeps working. The really cool thing is that it applies to anything called avconv, for as long as it (cpulimit) is running – so even though my batch conversion script kicks off many copies of avconv (sequentially), all of them are tamed by one instance of cpulimit:

cpulimit -e aconv -l 45 > /dev/null 2>&1

Not only that, if there is nothing called avconv running, cpulimit will wait until there is – then control it. This means I can put the above invocation into my script anywhere before the loop that starts avconv; cpulimit will start up and wait until avconv starts, then limit it.

The same applies if you specify a process ID to cpulimit – if the process ID doesn’t yet exist, cpulimit will wait until it does. In short, you don’t have to apply cpulimit to a running process, which in turn means that you don’t have to figure out process IDs in order to use it.

Because cpulimit will wait indefinitely for  process to start, it’s important to kill it when it is no longer needed, especially if you invoked it on a PID, because that PID may be re-used after a while. In my scripts, the code looks something like this:

   ...
   cpulimit -e aconv -l 45 > /dev/null 2>1
   CPULIMITPID=$!
   ...
   kill -TERM $CPULIMITPID
   ....

cpulimit doesn’t seem clever enough to handle multiple simultaneous instances though. When I start a second instance of avconv, it quickly runs up to 110% of CPU or more, while the first instance remains at around 45%. Possibly in such cases the specific PIDs need to be controlled. But it’s still very useful.

cpulimit – highly recommended!