[Nitro] Nitro leaking?

[Jonathan Buch]

Hi,

what started as the question of some benchmarks, showing if Rails or Nitro  
is faster in rendering a simple app, I guess we found quite some leak.

Rails benchmark on which the Nitro test is modelled from:
http://blog.kovyrin.net/2006/08/28/ruby-performance-results/

[19:58] 	<wyhaines>	I just did a fresh install of nitro via gem install,  
then ran the code snippet that you sent to me.  Process size grows quite a  
lot with every 10000 requests I push through it.
[19:58] 	<wyhaines>	On both Ruby 1.8.4 and 1.8.5.
[20:04] 	<Vagabond>	might be something to toss the mailing list's way
[20:04] 	<Vagabond>	how much are we talking here BTW?
[20:05] 	<wyhaines>	Process size, for what it's worth, was close to 30Mb  
after 10k requests.
[20:05] 	<wyhaines>	Eh, I closed the window, but I think it was about 60Mb  
after 30k requests.
[20:06] 	<Vagabond>	ugh, yeah that's not good
[20:08] 	<Kashia>	heavy :)
[20:13] 	<wyhaines>	And, for what it's worth, I was running this on a  
version of Ruby patched to fix the leak in Hash that I found at the end of  
last week.
[20:14] 	<wyhaines>	And it was running with a modified Mutex, too, that  
eliminates the potentially troublesome interaction with array.c's internal  
memory management, so it's not related to that, either.
[20:14] 	<wyhaines>	And the leak was about the same in both webrick and  
mongrel.

Jo
PS:  As consolation, Nitro was found to be way faster.. :P

-- 
Feel the love
http://pinkjuice.com/pics/ruby.png
-------------- next part --------------
A non-text attachment was scrubbed...
Name: benchmark.zip
Type: application/zip
Size: 1666 bytes
Desc: not available
Url : http://rubyforge.org/pipermail/nitro-general/attachments/20060905/82b19082/attachment.zip