Performance tuning...

[appel <ar...@siminn.is>]

Hi

I am in need of performance tuning my Geronimo (Jetty) server to accommodate
for a specific type of load burst.


The behavior I need to tune Geronimo to handle goes something like this:

- Every minute, 3000 unique clients will load up a set of 4 pages.
- This happens every minute for 10 minutes, so all in all there are 30.000
unique clients that will try to connect and request those 4 pages over a
period of 10 minutes.


The pages are very plain, two html pages, two jsp pages that do minimum
stuff.

The bottleneck here is the number of clients.

How can I performance tune Geronimo (Jetty container) to handle this type of
high client load?

(I've already increased the max-threads to 500, but when I do performance
load-test in JMeter I get lots of errors.)

Thanks.
-- 
View this message in context: http://www.nabble.com/Performance-tuning...-tp24320965s134p24320965.html
Sent from the Apache Geronimo - Users mailing list archive at Nabble.com.

Re: Performance tuning...

[Gianny Damour <gi...@optusnet.com.au>]

Hi,

I believe you should deploy an HTTP cache in front of Geronimo. You  
will observe way better performance improvements using this approach  
than by trying to tune Geronimo. You may want to have a look at Squid:

http://www.squid-cache.org/


If your two JSP pages are client specific, split them into client  
specific and client agnostic fragments and use a caching engine to  
cache the client agnostic fragments.

Thanks,
Gianny

On 03/07/2009, at 8:31 PM, appel wrote:

>
> Hi
>
> I am in need of performance tuning my Geronimo (Jetty) server to  
> accommodate
> for a specific type of load burst.
>
>
> The behavior I need to tune Geronimo to handle goes something like  
> this:
>
> - Every minute, 3000 unique clients will load up a set of 4 pages.
> - This happens every minute for 10 minutes, so all in all there are  
> 30.000
> unique clients that will try to connect and request those 4 pages  
> over a
> period of 10 minutes.
>
>
> The pages are very plain, two html pages, two jsp pages that do  
> minimum
> stuff.
>
> The bottleneck here is the number of clients.
>
> How can I performance tune Geronimo (Jetty container) to handle  
> this type of
> high client load?
>
> (I've already increased the max-threads to 500, but when I do  
> performance
> load-test in JMeter I get lots of errors.)
>
> Thanks.
> -- 
> View this message in context: http://www.nabble.com/Performance- 
> tuning...-tp24320965s134p24320965.html
> Sent from the Apache Geronimo - Users mailing list archive at  
> Nabble.com.
>

Re: Performance tuning...

["viola.lu" <vi...@gmail.com>]

Any log about your errors? And some suggestions: maximize your OS tcp
connections and open file descriptors, on linux:
modify /etc/sysctl.conf with net.ipv4.ip_local_port_range = 1024 65000 ,this
will promise about 6000 tcp connections and run sysctl -p to check whether
this setting works fine.

modify /etc/security/limits.conf with 
root soft nofile 10240 #you can change 10240 as you would like
root hard nofile 10240

edit /etc/pam.d/login with
session required pam_limits.so

Pls try it!


appel wrote:
> 
> Hi
> 
> I am in need of performance tuning my Geronimo (Jetty) server to
> accommodate for a specific type of load burst.
> 
> 
> The behavior I need to tune Geronimo to handle goes something like this:
> 
> - Every minute, 3000 unique clients will load up a set of 4 pages.
> - This happens every minute for 10 minutes, so all in all there are 30.000
> unique clients that will try to connect and request those 4 pages over a
> period of 10 minutes.
> 
> 
> The pages are very plain, two html pages, two jsp pages that do minimum
> stuff.
> 
> The bottleneck here is the number of clients.
> 
> How can I performance tune Geronimo (Jetty container) to handle this type
> of high client load?
> 
> (I've already increased the max-threads to 500, but when I do performance
> load-test in JMeter I get lots of errors.)
> 
> Thanks.
> 

-- 
View this message in context: http://www.nabble.com/Performance-tuning...-tp24320965s134p24321181.html
Sent from the Apache Geronimo - Users mailing list archive at Nabble.com.