From: Jacques Gelinas (jack_at_solucorp.qc.ca)
Date: Wed 24 Oct 2001 - 07:12:36 BST
On Mon, 22 Oct 2001 17:16:29 -0500, Sam Vilain wrote
> Perhaps your per-s_context ulimit should be thought of as a
> per-s_context "resources allocation". Resource allocation includes
> weighted CPU scheduling, weighted disk IO scheduling, network device
> bandwidth weighting (not necessarily in the same kernel data
> structure; this one is probably better off done per-IP with QoS),
> ulimits, quotas, etc.
Yes this is what I had in mind. On the scheduling side, all I wanted was
a way to limit a vserver so it won't take all the server to itself.
Note that you can still "nice" all process in the vserver and they can't
get more. So the flag is more a "per s_context" weight vs a per process
weight. The nice level is probably the thing to use to fine tune, although
it is uneasy to renice a whole bunch of moving process.
So yes, a weight instead of a flag could be used.
> ie, say you have two virtual servers - BATCH and INTERACTIVE. At 8pm,
> BATCH's disk scheduling priority is raised to 4, and INTERACTIVE's to
> 1. So BATCH gets 80% of the disk time. Perhaps you still want
> INTERACTIVE to have lots of CPU, so you set BATCH's CPU scheduling
> priority to 3, and INTERACTIVE's to 7 at the same time.
This makes sens
---------------------------------------------------------
Jacques Gelinas <jack_at_solucorp.qc.ca>
vserver: run general purpose virtual servers on one box, full speed!
http://www.solucorp.qc.ca/miscprj/s_context.hc