From: Bob Aman <vacindak@...>
Date: 2006-09-05T07:32:03+09:00
Subject: Re: best-performing Rss parser

> Just wanted to give everyone who's curious an update on this situation.
>
> I ended up comparing just Feedtools and Syndication, since the others
> were not as feature-complete as I would have liked.
>
> In our system tests, Syndication performs a lot better than Feedtools,
> but I won't publicize the results here since those are specific to our
> system.
>
> In just the parsing portion of parsing
> http://www.digg.com/rss/chandrasonic/index2.xml (grabbing 40 entries, 40
> titles, 40 dates, etc etc) Syndication took 0.062467 seconds of total
> time and Feedtools took 4.227067 seconds of total time.  This level of
> performance difference is highly reproduceable and I am using this
> specific feed just to show some numbers.
>
> Syndication does have some downsides however.  The programmer must
> specifically whether a feed is RSS or Atom since Syndication has no
> built-in distinguishing mechanism.  Syndication also seems to have
> incomplete time-stamp support.  Slashdot time stamps, for example,
> doesn't work for me.
>
> Disclaimer: I'm not an expert on Ruby, RSS feeds or benchmarking, but
> just wanted to share my results.  Thanks Bob for all your help along the
> way.

The results aren't even remotely suprising to me, but yeah, they
really underscore several points:

1) REXML is a performance stinker (and by extension, FeedTools 10x more so)
2) If performance is an issue, a parse-at-all-costs parser is going to
be a problem unless it's written in C or something similarly fast
3) As always, use the right tool for the job -- FeedTools isn't going
to scale, so if you need scalability, don't use it
4) If you're dealing with ~500 feeds or less, you want to always get
the right answers back, you don't mind the size of the library, and
you want a Ruby-only solution, FeedTools is what you want

In case it's not obvious, point #4 has a lot of conditions.

At some point in the distant future, I intend to write a pure C
library that -will- scale, but that's a long way off, because for the
forseeable future I'm going to be working on GentleCMS.

Cheers,
Bob Aman
--
AIM: sporkmonger
Jabber: sporkmonger@jabber.org