Subject: Re: 1:1 thread model
To: None <tech-kern@netbsd.org>
From: Gary Thorpe <gathorpe79@yahoo.com>
List: tech-kern
Date: 01/10/2003 14:03:55
Mail archive email which has 1:1 vs. M:N comparison in Linux (NPT vs.
NPGT):
https://listman.redhat.com/pipermail/phil-list/2002-September/000009.html

Note: no details of how the testing was done are provided or even
source code for the testing. Both seems to have several versions, but
no details are provided as to which version of each is compared.

Some other notes on the area:
http://www.kegel.com/c10k.html#threaded

NPGT home page and performance comparisons:
http://www-124.ibm.com/pthreads/
http://www-124.ibm.com/pthreads/docs/pingpong.2way.pdf
http://www-124.ibm.com/pthreads/docs/pingpong.4way.pdf [the link for
this in the page is wrong, this is correct]

Now, according to these tests, NPT is faster when the parallelism is
lower but it doesn't scale as well on the 4-way machine (which is what
I gather is generally predicted for 1:1 vs. M:N). Hence, the NGPT page
claims that NPGT is the fastest and most scaleable one for Linux. Since
the NPT version is so new (and probably not as tested/debugged) there
is still some possibility it could be better eventually.

Everyone seems to have data showing theirs is better though.....

______________________________________________________________________ 
Post your free ad now! http://personals.yahoo.ca