On Wed, 13 Feb 2013, Bob Friesenhahn wrote:

On Wed, 13 Feb 2013, Jim Klimov wrote:

For clarity of experiment, it might help to disable cpupm completely
and/or to run the experiment with a larger threshold (and loop more
of your program's executions).

I disabled CPU power-management and cold-booted and this is what 'time' reports for successive runs with a 1 second sleep in between. Notice no more bouncy total time readings but also performance still seems inadequate (compared with Linux) with excessive system time. The total time reading is similar to the better readins when results were inconsistent.

I am unable to find anything wrong with my system and in fact benchmark runs which run for a long time and re-use memory show a 2-2.5X performance advantage of the Illumos-based system over the Linux system (due to the CPU).

However, I did experiment and learned something significant:

% pagesize -a
4096
2097152
1073741824
% ptime /usr/bin/ppgsz -o heap=4096 ./magick large.gif

real        9.547871680
user        5.821892111
sys         2.812548350
% ptime /usr/bin/ppgsz -o heap=2097152 ./magick large.gif

real        6.883969267
user        4.295124634
sys         1.790358267
% ptime /usr/bin/ppgsz -o heap=1073741824 ./magick large.gif

real        7.978930675
user        4.610727541
sys         2.194304364

Notice that the real and system time were dramatically reduced by using a larger page size. Additional testing shows system time being cut in half and large reductions in reported user time as well.

Remember that the Linux total run time was 6.179 seconds so using the larger pages brings the OI result close to Linux (but it should be faster!).

It seems that Solaris 11 has made considerable changes in the kernel for large pages. Obviously, it can help user-space programs as well.

Bob
--
Bob Friesenhahn
[email protected], http://www.simplesystems.org/users/bfriesen/
GraphicsMagick Maintainer,    http://www.GraphicsMagick.org/


-------------------------------------------
illumos-discuss
Archives: https://www.listbox.com/member/archive/182180/=now
RSS Feed: https://www.listbox.com/member/archive/rss/182180/21175430-2e6923be
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=21175430&id_secret=21175430-6a77cda4
Powered by Listbox: http://www.listbox.com

Reply via email to