Managing R CPU/memory usage on Linux
Control groups is probably the way to go if you have it, as Etienne suggested, but also consider increasing swappiness, e.g. echo 100 >> /proc/sys/vm/swappiness (or more likely the equivalent in whatever /etc/sysctl.conf is on Ubuntu). You are quite low on RAM. You will of course want a decent-sized swap file. And without control croups you should set resource limits, cf. man limits.conf. You could also try 'ionice -c 3' but the benefit is likely to be low (nice -19 already drops your scheduling class to the equivalent of -c 2 -n 7). Allan
On 19/11/10 21:41, Davor Cubranic wrote:
We have some Linux compute servers that our users occasionally overwhelm with their R batch jobs to the point where the machines are completely unresponsive and have to be rebooted. This seems to be happening more often recently, and got me wondering what other people do to manage the CPU/memory resources used by R on their servers. We 'nice -19' the R process, but that doesn't seem to help. Are there any other R options or OS settings that would be useful? Or should I consider installing a queuing manager and closing the servers to interactive logins? As far as I can tell, out users just run existing R packages from CRAN, and there is no parallelization or distributed computing going on. The machines are dual-CPU 64-bit Intels with 4GB of RAM and running Ubuntu 8.04. So they won't be making the TOP500 list any time soon, but I would have hoped the kernel would be a litter better at squelching down CPU and memory hogs. Thanks, Davor
_______________________________________________ R-sig-hpc mailing list R-sig-hpc at r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-hpc