Reducing thrashing effect in linux

[atoka wotsa]

Can we have a Medium term scheduler in linux to reduce thrashing effect, Medium term scheduler will group the processes such that the memory requirement of each group is almost close to the physical memory available and schedule them. the ordinary scheduler will then schedule the processes within the group.


kindly give your comments and suggestions...


---------------------------------
All-new Yahoo! Mail - Fire up a more powerful email and get things done faster.

[James Stevenson]

Its not required on a correctly configured system. You can limit the amount
of memory that can be used by users / tasks.

If this is still a problem then you are probably running something on a
machine that cannot run the tasks you require it to.



There was a version I did see some time ago that when the scheduler thought
the system was thrashing it would pause tasks for a longer period of time to
allow the system to recover from a load spike. But it would not work for a
system trying to sustain a heavy load.





_____

From: kernelnewbies-***@nl.linux.org
[mailto:kernelnewbies-***@nl.linux.org] On Behalf Of atoka wotsa
Sent: 13 September 2006 07:30
To: ***@nl.linux.org
Subject: Reducing thrashing effect in linux



Can we have a Medium term scheduler in linux to reduce thrashing effect,
Medium term scheduler will group the processes such that the memory
requirement of each group is almost close to the physical memory available
and schedule them. the ordinary scheduler will then schedule the processes
within the group.





kindly give your comments and suggestions...



_____

All-new
<http://us.rd.yahoo.com/evt=43256/*http:/advision.webevents.yahoo.com/mailbe
ta> Yahoo! Mail - Fire up a more powerful email and get things done faster.

[Rik van Riel]

Post by atoka wotsa
Can we have a Medium term scheduler in linux to reduce thrashing effect,
Medium term scheduler will group the processes such that the memory
requirement of each group is almost close to the physical memory
available and schedule them. the ordinary scheduler will then schedule
the processes within the group.

You could create one. However, second level scheduling is
going to be less and less feasible over time.

The reason is that tasks grow with memory size, not with
disk speed. The time it takes to write all of memory to
disk has grown from 5 seconds in 1992 or so to several
minutes today. The time needed to swap processes in and
out has grown similarly.

You may be better off reducing the thrashing effect by
further tuning the swap token code which the 2.6 kernel
already has.

For more information on the challenges there, please see:

http://linux-mm.org/SwapTokenTuning

Success! ;)