Forum: Ruby on Rails Performance of to_xml, versus to_json

Announcement (2017-05-07): www.ruby-forum.com is now read-only since I unfortunately do not have the time to support and maintain the forum any more. Please see rubyonrails.org/community and ruby-lang.org/en/community for other Rails- und Ruby-related community platforms.
Eric Mill (Guest)
on 2007-05-02 00:45
(Received via mailing list)
I've been doing some benchmarks on how fast ActiveRecord's #to_xml is
on large datasets, versus the speed of A
ActiveRecord's #to_json, and the difference is rather alarming.
#to_xml becomes quickly unusable for large datasets, while to_json
remains tolerable.

The important thing here isn't how fast #to_json is -- it's that you
can't practically use #to_xml anymore after a certain point, a low
point.

Some basic benchmarks, with a User model that has 5 fields: an
integer, a string, a text, and two datetimes.

Commands benchmarked:
  to_xml - "User.find(:all).to_xml"
  to_json - "User.find(:all).to_json"

50 items:
  to_xml - ~0.20 seconds
  to_json - ~0.02 seconds

500 items:
  to_xml - ~6.9 seconds
  to_json - ~0.4 seconds

1000 items:
  to_xml - ~25 seconds
  to_json - ~0.9 seconds

8000 items:
  to_xml - >10 minutes (didn't wait for it finish)
  to_json - 10.7 seconds

Can anybody comment on this?  With Rails pushing ActiveResource and
REST like crazy, services which really start allowing people to list
their products and userbases in XML are going to have a tough time
with out-of-the-box Rails, with performance like this.

-- Eric
Rick O. (Guest)
on 2007-05-02 04:46
(Received via mailing list)
> Can anybody comment on this?  With Rails pushing ActiveResource and
> REST like crazy, services which really start allowing people to list
> their products and userbases in XML are going to have a tough time
> with out-of-the-box Rails, with performance like this.

From your numbers, this isn't an issue for the vast majority of
services.  I'm not sure why anyone would send 8000 items in a request.
 But, if this concerns you, I'd be happy to commit some performance
patches for it.

--
Rick O.
http://lighthouseapp.com
http://weblog.techno-weenie.net
http://mephistoblog.com
Chad P. (Guest)
on 2007-05-02 05:41
(Received via mailing list)
> From your numbers, this isn't an issue for the vast majority of
> services.  I'm not sure why anyone would send 8000 items in a request.
>  But, if this concerns you, I'd be happy to commit some performance
> patches for it.

Hi Rick.  Yeah, Eric's 8000 example is the extreme case, but any
performance patches would be great, and we're willing to help in any
way we can.

I was also wondering what to_json wasn't doing (besides the
obvious :include stuff) that it should be doing to bring it on par
with the to_xml code.  We'll likely switch to json for our specific
application, but it'd be good to know any issues before we dive in
(granted, we're more than willing to fix any issues we find once we
get in there).

thanks,
-Chad
Rick O. (Guest)
on 2007-05-02 05:54
(Received via mailing list)
On 5/1/07, Chad P. <removed_email_address@domain.invalid> wrote:
> I was also wondering what to_json wasn't doing (besides the
> obvious :include stuff) that it should be doing to bring it on par
> with the to_xml code.  We'll likely switch to json for our specific
> application, but it'd be good to know any issues before we dive in
> (granted, we're more than willing to fix any issues we find once we
> get in there).

Eric's email sounded kind of like he was volunteering.  And, I am a
very willing receiver of such patches :)

I've done a bit of json hacking using the fjson gem:
http://svn.techno-weenie.net/projects/plugins/json...

The big problem I ran into was worrying about dates and times.  My
plugin converts them all to xmlschema times.  But, there's no way to
tell the type of incoming objects, so I used a regex to cast xmlschema
times back to Time objects.  This could cause problems if someone
posted an xmlschema time as the body of a post though.  Perhaps a
better solution is to assume *_at/*_on fields are times?  I'm not
sure.  I'd like to stay consistent with other json API's, but from
what I gathered they differed anyway.  So, I shelved the plugin for
now.

--
Rick O.
http://lighthouseapp.com
http://weblog.techno-weenie.net
http://mephistoblog.com
This topic is locked and can not be replied to.