With the new controllers in CKRM in place, has anyone looked at whether these controllers are sufficient to isolate "anti-social" behavior (e.g., fork bombs, i/o bombs, etc.)? In other words, assuming these anti-social processes were placed into an appropriate class, do the other classes obtain their share of the system resources? I ask as this is an important piece that we wish to get right for PlanetLab.
Marc
At OLS, with e15 + a bug fix (included in e16), we'd successfully run
a fork bomb that got choked by the numtasks controller at whatever process limit we set - rest of the system was responsive.
Also, a cpu hog was run with some limits and the foreground activity's /interactivity looked good. Don't remember how closely the shares of the non-hog processes were respected but from Haoqiang's experiments (which run groups of cpu bound processes), they were quite accurate.
Memory testing for CKRM typically involves memory hogs since we regulate mem only under overload and earlier results have shown that such hogs are contained well provided there isn't shared memory involved.
I haven't run any i/o bombs on the iosched yet.
Regards, Shailabh
------------------------------------------------------- This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170 Project Admins to receive an Apple iPod Mini FREE for your judgement on who ports your project to Linux PPC the best. Sponsored by IBM. Deadline: Sept. 24. Go here: http://sf.net/ppc_contest.php _______________________________________________ ckrm-tech mailing list https://lists.sourceforge.net/lists/listinfo/ckrm-tech
