146343 - optimize the pthread/(solaris thread) to enhance the performance of netscape on solaris.

Status: RESOLVED INVALID

(Core :: XPCOM, defect)

Description

[jack.jia]

Solaris being a truely MP/MT systems can slow some apps that do some specific 
stuff to sync between threads.

Comment 1

[Henry Jia]

Confirm the work. Change the status to "New".

Comment 2

[Bradley Baetz (:bbaetz)]

-> nspr

browser general doesn't fix bugs. 

Comment 3

[Wan-Teh Chang]

Could you be more specific about what performance enhancements
you want in NSPR?

Comment 4

[Nidheesh Dubey]

I think the original wordings of this bug are not very clear about the suspected
issue. Here is my take on it. First of all this is exploratory work and we are
not concluding that it is indeed a real issue at the moment. On solaris we find
some issues for example that might be related to problems in specific use of threads
1. Excessive polling bug 86632 got fixed by making timer thread safe on linux
but not on solaris. We still find lot of system activity on M1.0 or Netscape7pr1
on solaris.
2. In doing performance analysis of Siebel's static pages we found that there is
a fixed delay of 1-1.5 seconds on solaris. Since Mozilla/Netscape uses threads
any sync effort between threads that uses fixed delays could possibly be the
culprit.
3. We also would also like to explore if doing disk read/write for cache can be
done from a separate thread as disk cache read/write/stats seem to be quite time
consuming and seems to be causing a delay on worker thread that is drawing the
web page.
4. Bug 123920 found at Siebel where the gc was dumping core due to MP nature of
java and netscape on solaris (two separate processes)

Comment 5

[jack.jia]

   The patch for bug 86632 is too old and it isn't checked in the trunk, but I 
think the patch for bug 78611 can resolve the excessive polling. with it, the 
polling & ioctl became much fewer than before. I had checked it comparing with 
the mozilla094 without the patch.

   

Comment 6

[Wan-Teh Chang]

1. This bug needs to be reassigned to a more appropriate
product/component.  None of the bugs cited so far are
NSPR bugs.  I am tentatively changing the product/component
to Browser/Threading.

2. The summary and description of this bug are so general
that nobody will be able to honestly mark the bug fixed.
I suggest that you open new bugs on specific performance
problems of Netscape on Solaris (such as the ones described
in comment #4), and resolve this bug invalid.

Comment 7

[Hixie (not reading bugmail)]

Moving all threading bugs to XPCOM. See bug 160356.

Comment 8

[Wayne Mery (:wsmwk)]

Reporter is MIA. Based on #2 of comment #6 it is reasonable to close this bug.  

Feel free to open new bugs against specific fixable issues as suggested, or take
up the cause on the ~40 core bugs that mention solaris.