2003-11-14 05:14:05

by Roger Larsson

[permalink] [raw]
Subject: [CODE RFC] redefining sched_setscheduler using LD_PRELOAD

Hi,

(Jack O'Quin got me thinking about rt_monitor again - thanks!)

This is an alternative to capabilities / SCHED_SOFTRR...

I use my old rt_monitor
* protects the system on RT overload
* performs the actual setting of scheduler type
* it does not even have to be visible for lusers!

My old testprogram RT has been converted to
a library fit for preloading.

It works like this.

As root start
# rt_monitor

As normal user start latencytest
$ ./rts path/to/latencytest none 2 128 10

rts has two options
-i do not raise the priority.
-v be more verbose.
* It does also limit the effective min/max range.

Do also try (don't try this as root unless rt_monitor is running)
$ ./rts path/to/latencytest none 2 1024 99
* In this case rt_monitor captures the process and renices it.
(SCHED_OTHER + nice)

BUGS:
* better build and installation system
* could ignore RT processes already running when rt_monitor starts
* rt_monitor functionallity has not been reverified...
* it can't lock memory :-( No PIDs in calls.
* code that checks uid before using the redefined functions won't work
(I have an modified latencytest)
* client get stuck if there is no monitor running.

/RogerL

--
Roger Larsson
Skellefte?
Sweden





Attachments:
(No filename) (1.20 kB)
rts.zip (15.52 kB)
Download all attachments