[lkml]   [2004]   [Mar]   [4]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRE: [RFC][PATCH] O(1) Entitlement Based Scheduler
At a previous employer (so code not available) I used a simple expedient to
solve this very problem. I had a custom program "shim.c" that tweaked
priorities and environment variables. Basically a fistful of lines that
would take argv[0], look for the file named ".shim_"+basename(argv[0]) {in a
well-defined location} to load some simple environment and path and priority
overrides, apply these changes and then setuid itself back to the real user
and exec() the real program with the received args.

It had some few degenerate cases (shmming out from under a setuid program
was the primary one) but it worked out rather well and had little-to-no
meaningful overhead.

You set up a /usr/local/bin (or equivalent) directory, link shim into that
directory named as the various programs that need to be boosted (e.g. xmms
etc) and put that directory earlier on the path than the real executable.
{If this directory only contains shims, it is useful to code shim.c to
remove that directory from the PATH.)

This technique lets the administrator have fine-grained control of a
reasonable list of priority promotions and permissions overrides without
having to move anything into kernel space or running status daemons.


I would think that while fork() should keep the heuristics of its parent,
exec() would probably need to do some normalizing.


Has this scheduler been tried for applications like CD burning?


-----Original Message-----
[] On Behalf Of Peter Williams
Sent: Wednesday, February 25, 2004 6:18 PM
To: Timothy Miller
Subject: Re: [RFC][PATCH] O(1) Entitlement Based Scheduler

Timothy Miller wrote:
> <snip>
> In fact, that may be the only "flaw" in your design. It sounds like
> your scheduler does an excellent job at fairness with very low overhead.
> The only problem with it is that it doesn't determine priority
> dynamically.

This (i.e. automatic renicing of specified programs) is a good idea but
is not really a function that should be undertaken by the scheduler
itself. Two possible solutions spring to mind:

1. modify the do_execve() in fs/exec.c to renice tasks when they execute
specified binaries
2. have a user space daemon poll running tasks periodically and renice
them if they are running specified binaries

Both of these solutions have their advantages and disadvantages, are
(obviously) complicated than I've made them sound and would require a
great deal of care to be taken during their implementation. However, I
think that they are both doable. My personal preference would be for
the in kernel solution on the grounds of efficiency.

Dr Peter Williams, Chief Scientist
Aurema Pty Limited Tel:+61 2 9698 2322
PO Box 305, Strawberry Hills NSW 2012, Australia Fax:+61 2 9699 9174
79 Myrtle Street, Chippendale NSW 2008, Australia

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

 \ /
  Last update: 2005-03-22 14:01    [W:0.181 / U:3.680 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site