gnu-arch-users
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Gnu-arch-users] archzoom


From: Mikhael Goikhman
Subject: Re: [Gnu-arch-users] archzoom
Date: Wed, 18 Oct 2006 20:40:36 +0000
User-agent: Mutt/1.5.11

[I was offline for a long time, so the late answer.]

It does not seem the "bad performance of archzoom" issue is presented
correctly. I can't say whether the Savannah administrator ever enabled
archzoom. As for gna.org, the issue looks pretty simple. The
administrator decided he has no many servers and wants to dedicate his
single server to subversion browsers and not to arch.

Here is some info that may help. The default archzoom configuration is
conservative in that it does not try to write things to a user's disk.
This is good for a demo installation, but requires tuning for production.
The ArchZoom FAQ explains how to trivially define revision library that
may cause tla to be drastically faster. It also explains how to keep the
size of this revision library constant using "axp revlib prune --params".

Another related problem is that many developers have unresponsively huge
branches of thousands of revisions without a single cacherev, that makes
certain tla operations totally stuck when working on these branches. A
policy of automatic creation a cacherev every 50 revisions partially
solves this problem.

By default, archzoom forbids search engines to crawl its pages, but some
web server misconfiguration or misbehaved robots may cause some unwanted
bombing too. To solve this problem archzoom has a number of configuration
options, for example it may limit the number of its instances. Problems
like these described in the last 3 paragraphs are the real issue and it
has nothing to do with archzoom.

As a side note, certain tla operations (see threads about "tla abrowse",
for example) are needlessly unoptimal, and may be improved noticeably.
Many tla commands miss a limit, as others mentioned.

I think the perl layer has a quite little overhead that is minor compared
to the real problems related to tla and a web server. I don't really see
"tla librification" essential, as long as tla behaves like a good library
(and it is pretty close to it); fork+system is cheap on unix.

As for the idea of making tla to behave more in a daemon manner, I am a
bit skeptical about this, but I should see more info to form any opnion.
ArchZoom implements some of this stuff, it has own mechanism of caching
many of the views for several hours, so that tla is not called. I think
it may also work under mod_perl if one wants this, however the standard
CGI mode together with limiting the number of archzoom instances may work
as well.

Regards,
Mikhael.




reply via email to

[Prev in Thread] Current Thread [Next in Thread]