You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by plasma <pl...@pchome.com.tw> on 2003/07/17 15:35:31 UTC
Slowness of svn and http
Hi all,
Today, I set up a svn network server, import source of mozilla 1.0,
then check out.
........
.............
......................
After one hour, I was still waiting checkout to finish. Isn't that
too slow to be useful?
I'll try svnserve tomorrow, see if it's better.
plasma
==========================================================
�A��d,VISA���A�R��
http://edm-prg.epaper.com.tw/click.php?ad_code=15650
==========================================================
PChome�u�W�ʪ��g�~�y�G��T���BDV�ѤѰe
http://shopping.pchome.com.tw/
==========================================================
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Slowness of svn and http
Posted by plasma <pl...@pchome.com.tw>.
On Thu, Jul 17, 2003 at 10:48:21AM -0500, Ben Collins-Sussman wrote:
> plasma <pl...@pchome.com.tw> writes:
>
> > Today, I set up a svn network server, import source of mozilla 1.0,
> > then check out.
> >
> > After one hour, I was still waiting checkout to finish. Isn't that
> > too slow to be useful?
>
> What metric are you using to measure speed? What is the definition of
> "too slow"? The mozilla source is a *huge* dataset. How do you know
> one hour is "too slow"?
>
> Repeat the experiment with a cvs repository, compare the results, and
> I'll listen to you. I want science here, not fuzzy feelings. :-)
Sussman said I need to be scientific, so here comes the result.
The server is FreeBSD 4.7. The client is my IBM X23, FreeBSD 4.8.
I import mozilla 1.0's source, which is extracted from
mozilla-source-1.0.tar.bz2, into a repository, then check out the
source over LAN.
1. svn co http://$SVN_ROOT/mozilla-1.0 mozilla
172.907u 129.123s 1:54:06.76 4.4% 104+26171k 24311+933io 13pf+0w
2. svn co http://$SVN_ROOT/mozilla-1.0 mozilla
174.115u 130.746s 1:57:59.33 4.3% 104+26385k 23718+933io 5pf+0w
3. svn co svn+ssh://$SVN_ROOT/mozilla-1.0 mozilla
166.331u 109.494s 50:13.02 9.1% 104+24533k 23261+926io 0pf+0w
4. cvs -d :ext:$CVS_ROOT checkout mozilla
57.354u 48.390s 3:07.65 56.3% 204+1037k 269+430io 1pf+0w
The difference between 1 and 2 is that 2 is invoked with
"http-compression = no" at client side.
As you can see, svn+http takes almost 2 hours to finish. svn+ssh is
much better, 50 minutes. cvs+ssh just takes 3 minutes.
I think I can say "It's slow" aloud now. :)
plasma
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Slowness of svn and http
Posted by Greg Stein <gs...@lyra.org>.
On Thu, Jul 17, 2003 at 10:49:01AM -0700, Jack Repenning wrote:
> At 12:28 PM -0500 7/17/03, Kirby C. Bohling wrote:
> >As nearly as I can tell, that's roughly what a check out of HEAD is.
> >You get the files from apache, you uncompress them once, you write them
> >to disk twice. That'd give you a rough idea of what the overhead of
> >running subversion is. Did I get that correct?
>
> I'm not sure what this experiment tells us. The net result of this
> process is only files on your local disk; it does not connect you to
> the VC system. If that's all you need, then by all means, just
> download and unpack the tarball once!
>
> On the other hand, if you're actually looking for VC (so you can get
> changes more efficiently, or so you can commit changes back in, or
> all the other stuff that VC provides), then it's hard to understand
> the spread between this experiment and actual VC. This experiment
> doesn't include creating the various state stuff in .svn/ (you've
> covered the base text copy, but not the rest). This experiment also
> doesn't cover any of the server-side work necessary to accomplish
> your request (selecting the particular versions you want).
Right, but the experiment tells us about the resulting costs of those
overheads. If we spend another 30 minutes over {download, unzip, untar,
untar}, then it says we spend 30 minutes creating .svn and with server-side
work, or with wire overhead, or whatever.
In an ideal world, a checkout would be just as fast as a download. So the
question is: how close are we?
Cheers,
-g
--
Greg Stein, http://www.lyra.org/
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Slowness of svn and http
Posted by Jack Repenning <jr...@collab.net>.
At 12:28 PM -0500 7/17/03, Kirby C. Bohling wrote:
>As nearly as I can tell, that's roughly what a check out of HEAD is.
>You get the files from apache, you uncompress them once, you write them
>to disk twice. That'd give you a rough idea of what the overhead of
>running subversion is. Did I get that correct?
I'm not sure what this experiment tells us. The net result of this
process is only files on your local disk; it does not connect you to
the VC system. If that's all you need, then by all means, just
download and unpack the tarball once!
On the other hand, if you're actually looking for VC (so you can get
changes more efficiently, or so you can commit changes back in, or
all the other stuff that VC provides), then it's hard to understand
the spread between this experiment and actual VC. This experiment
doesn't include creating the various state stuff in .svn/ (you've
covered the base text copy, but not the rest). This experiment also
doesn't cover any of the server-side work necessary to accomplish
your request (selecting the particular versions you want).
--
-==-
Jack Repenning
CollabNet, Inc.
8000 Marina Boulevard, Suite 600
Brisbane, California 94005
o: 650.228.2562
c: 408.835-8090
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Slowness of svn and http
Posted by "Kirby C. Bohling" <kb...@birddog.com>.
On Thu, 2003-07-17 at 10:48, Ben Collins-Sussman wrote:
> plasma <pl...@pchome.com.tw> writes:
>
> > Today, I set up a svn network server, import source of mozilla 1.0,
> > then check out.
> >
> > ........
> >
> > .............
> >
> > ......................
> >
> > After one hour, I was still waiting checkout to finish. Isn't that
> > too slow to be useful?
>
> What metric are you using to measure speed? What is the definition of
> "too slow"? The mozilla source is a *huge* dataset. How do you know
> one hour is "too slow"?
>
> Repeat the experiment with a cvs repository, compare the results, and
> I'll listen to you. I want science here, not fuzzy feelings. :-)
>
> (P.S. Yes, in general, we already know that our DAV network layer is
> slower than CVS. This isn't news. And we've discussed many times why
> this is so, and how to improve it. But mailing statements like "hey,
> this seems slow to me" is neither interesting news nor productive.)
Why not measure it against the yard stick of downloading the from the
Apache server .tar.gz from the same source (over the same link). The
ungzipping the tar, then untarring it twice. Maybe you need to have it
setup so that Apache does the zipping of the mozilla source, rather then
have it pre-zipped, then just untar it twice. Maybe no gzipping at all
to avoid the mod_gzip issues Ben just figured out.
As nearly as I can tell, that's roughly what a check out of HEAD is.
You get the files from apache, you uncompress them once, you write them
to disk twice. That'd give you a rough idea of what the overhead of
running subversion is. Did I get that correct?
If it takes 5 minutes to do that, and an hour to check out of
subversion, that says subversion has plenty of overhead on initial
checkout. If it takes 54 minutes to do it, and check out of subversion
is still an hour, that says something else, the subversion has about
~10% overhead. Something large like Mozilla would amortize the startup
costs for each method too so you could get an average for the streaming
checkout by size.
Kirby
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
> For additional commands, e-mail: dev-help@subversion.tigris.org
>
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Slowness of svn and http
Posted by Ben Collins-Sussman <su...@collab.net>.
plasma <pl...@pchome.com.tw> writes:
> Today, I set up a svn network server, import source of mozilla 1.0,
> then check out.
>
> ........
>
> .............
>
> ......................
>
> After one hour, I was still waiting checkout to finish. Isn't that
> too slow to be useful?
What metric are you using to measure speed? What is the definition of
"too slow"? The mozilla source is a *huge* dataset. How do you know
one hour is "too slow"?
Repeat the experiment with a cvs repository, compare the results, and
I'll listen to you. I want science here, not fuzzy feelings. :-)
(P.S. Yes, in general, we already know that our DAV network layer is
slower than CVS. This isn't news. And we've discussed many times why
this is so, and how to improve it. But mailing statements like "hey,
this seems slow to me" is neither interesting news nor productive.)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org