lkml.org 
[lkml]   [1998]   [Nov]   [17]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectMM observation (pathological swap using less than phys. ram)

Hello. This is not a "bug report," but rather an observation about
something a bit less than ideal, IMHO. I have a program called "hog"
which swallows a specified amount of memory, and then random accesses it
until terminated. This keeps the memory from "laying about" in swap! A
test I like to run I've named 'whomp.sh' which runs some number of 'hogs'
simultaneously. So here's the scenario:

I have 96MB ram, and hog configured to use 10M ram (10*1024*1024 as arg to
malloc). I ran 8 of these suckers, which should account for some 80MB of
ram used plus some overhead for the code, local variables etc. This test
holds true if run from a VC but I typically run it under an xterm with the
exact same results. The results:

Ultimately, all 8 'hogs' get arranged in physical memory. The process
counts 'random' accesses per second in each process and display this info,
so I can tell the instant all the ram is out of swap because the process
will go from about 75 accesses per second to 150000. So, after about 30
minutes all 8 are happily cranking away at the 100000 or so accseses per
second. But it takes 30 minutes of pathological swapping to get there.

This is the point: it takes 30 minutes of pathological swapping to arrange
80MB of ram into my 96MB of ram, but ultimately gets there!!!! Why?!?
Cross-referenced with 2.0.35: not only is the swapping way down (about 1
minute or 2), but I can actually run 9 'hogs' (90MB committed ram usage)
and see about 7 of them cranking away while 2 are in the mud.
2.1.129-pre5
cannot come close to this.

This has been since I caught up with 2.1 around 2.1.115, currently
2.1.129-pre5, on a PPro 200, 96MBram, gcc 2.7.2.3, swap on three IDE
disks.

If anyone would like to see my 'hog' let me know, it's as simple as you
expect...

David

--
/==============================\
| David Mansfield |
| david@cobite.com |
\==============================/


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 13:45    [W:0.031 / U:0.324 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site