Bug 151832 - CVS checkout really slow
Summary: CVS checkout really slow
Keywords:
Status: CLOSED NOTABUG
Alias: None
Product: Fedora
Classification: Fedora
Component: eclipse
Version: rawhide
Hardware: All
OS: Linux
medium
medium
Target Milestone: ---
Assignee: Ben Konrath
QA Contact:
URL:
Whiteboard:
Depends On:
Blocks:
TreeView+ depends on / blocked
 
Reported: 2005-03-22 18:52 UTC by Billy Biggs
Modified: 2007-11-30 22:11 UTC (History)
2 users (show)

Fixed In Version:
Clone Of:
Environment:
Last Closed: 2006-10-04 18:49:42 UTC
Type: ---
Embargoed:


Attachments (Terms of Use)
oprofile report (332.16 KB, text/plain)
2005-05-12 15:15 UTC, Anthony Green
no flags Details

Description Billy Biggs 2005-03-22 18:52:10 UTC
I am experimenting with using gcj/gij to run eclipse using the gcj packages from
the Fedora development repository.

One thing I have noticed when using gij or gcj-compiled eclipse is that CVS
checkouts seem really, really slow.  I am connecting to dev.eclipse.org
/home/eclipse as anonymous (pserver).  Is it possible that something at the
gcj/gij level is making network connections slow down?

Comment 1 Billy Biggs 2005-03-22 18:52:46 UTC
To clarify further, I am testing with eclipse I20050315-1100.

Comment 2 Andrew Overholt 2005-03-22 18:53:51 UTC
Moving to devel.  Assigning to me.

Comment 3 Andrew Overholt 2005-03-22 18:54:30 UTC
Really assigning to me.

Comment 4 Billy Biggs 2005-03-22 18:57:49 UTC
During checkout, the gij process is spinning the CPU.

Comment 5 Andrew Overholt 2005-03-22 19:00:52 UTC
This is CVS using ext, right?  It's over ssh?

Comment 6 Billy Biggs 2005-03-22 19:01:59 UTC
No, using pserver.

Comment 7 Anthony Green 2005-05-12 15:15:47 UTC
Created attachment 114297 [details]
oprofile report

I've attached an oprofile report of a pserver checkout of gnu-crypto.
I started profiling just as the cvs co started, and stopped as soon as it
finished.  
I haven't look at the report closely, other to note that we're spending the
vast majority of our time in the kernel - presumably waiting on stuff.	I also
don't understand why we're trying to load classes here.

Comment 8 Andrew Overholt 2005-05-18 13:41:01 UTC
I've been working on this 'cause it was really bugging me.  My test project was
gnu-crypto [1].  After fighting with oprofile and trying to figure out the cause
of the slow checkouts, I have now come to the conclusion that if we set the
compression level to 3 (or higher, I guess), checkout time is acceptable.  Since
I ran into similar problems with Sun's JVM (with gnu-crypto), I'm willing to bet
it's an upstream issue (or at least gnu-crypto exacerbates the issue).  I've
opened an Eclipse bug [2] to see if we can get the default compression set to 3
and if not, to see if there's a way we can set it in our builds.

[1]
:ext:anoncvs.org:/cvsroot/gnu-crypto
[2]
https://bugs.eclipse.org/bugs/show_bug.cgi?id=95697

Comment 9 Andrew Overholt 2005-06-06 15:11:06 UTC
I'd like to close this as NOTABUG because I get the same behaviour with our
native stuff as with Sun's JVM.

Thoughts?

Comment 10 Andrew Overholt 2005-10-11 11:21:12 UTC
I'm wondering if this was the same effect as bug 161483.  Billy/others, if you
get time, can you try the packages that will hit rawhide today (-15)?

Comment 11 Andrew Overholt 2005-10-20 15:28:46 UTC
I've noticed this myself now.  We do indeed take longer than the Sun JVM to
check things out of CVS regardless of compression level.

Comment 12 Andrew Haley 2006-03-13 13:28:38 UTC
Surely this one is now fixed.


Comment 13 Andrew Overholt 2006-03-13 15:04:47 UTC
I don't use the Sun JVM for much other than testing sometimes so I can't really
verify.  Ben, can you perhaps take a look at this?  Maybe we can get some data
out of the automated tests or something.

Comment 14 Igor Foox 2006-03-15 17:41:44 UTC
Blah blah.

Comment 15 Igor Foox 2006-03-15 17:42:03 UTC
(In reply to comment #10)
>I'm wondering if this was the same effect as bug 161483.  Billy/others, if you
>get time, can you try the packages that will hit rawhide today (-15)?
askldfjlsdf
sdfjsdfk


sfd

Comment 16 Ben Konrath 2006-10-04 18:44:46 UTC
I think we can close this bug now. Andrew, what do you think? 

Comment 17 Andrew Overholt 2006-10-04 18:49:42 UTC
Sure, and if checkout time bothers people, we can re-open.

Closing.


Note You need to log in before you can comment on or make changes to this bug.