Messages in this thread |  | | From | Jacques Gelinas <> | Date | Mon, 15 Oct 2001 21:43:31 -0500 | Subject | Re: Announce: many virtual servers on a single box (scheduling) |
| |
On Mon, 15 Oct 2001 17:22:43 -0500, James Sutherland wrote
> Hmmm - does this work with devfs?
Not tried. Sorry. Note that vserver can't mount.
> Have you looked at the "fairsched" patch for this? It seems to be > unmaintained since 2.4.0-testXX, but look close to your needs...
So far I am toying with a very simple solution. The current schedular (SCHED_OTHER) is using each process p->counter (time quantum used so far) as one data to compute the priority.
Security context are sharing a common struct context_info and it contains a counter too (atomic_t counter). When the process p->counter is decremented the context_info->counter is also decremented. This context_info counter is used along with the p->counter to compute the process priority. So if a vserver is running many processe, then their total cpu usage is accounted in their common context_info->counter, so they have lover priority (together) than a single process in another vserver.
Note that this idea is broad. A PAM module could be use to setup a per-user context_info, without switching the security ID and you end up with per-user fair schedular (well, per whatever).
Note also that this per context_info scheduling is a flag (you can only turn it on). By default, there is no special scheduling in a vserver.
--------------------------------------------------------- Jacques Gelinas <jack@solucorp.qc.ca> vserver: run general purpose virtual servers on one box, full speed! http://www.solucorp.qc.ca/miscprj/s_context.hc - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
|  |