Thanks for the quick responses.  We did run into some perm-gen problems
not too long ago and increasing the total perm gen memory did help.  

 

I will take a look at adding the GC vm arguments listed below and see if
I can find out just how much extra perm-gen each service is costing us -
this should be a good way to figure out a max # of services given our
hardware or the jvm's max memory limitations.  If I find anything
useful, I will post the results.

 

Thanks!

 

-Dave

 

From: Anthony Bull [mailto:[EMAIL PROTECTED] 
Sent: Monday, April 21, 2008 8:48 PM
To: [email protected]
Subject: Re: [Axis2] Performance and scalability for large numbers of
services

 

By the way, you can monitor your heap and permgen for your JVM with the
following VM arguments:

-verbose:gc -XX:+PrintClassHistogram -XX:+PrintGCDetails

every so often it will print out a line similar to below - note that on
my servers it prints about 1 per minute.

[Full GC [Tenured: 18012K->18038K(349568K), 0.6709127 secs]
20411K->18038K(519168K), [Perm : 36161K->36161K(36352K)], 0.6711622
secs]

As more web services come online or are hot re-deployed, the Perm part
will get bigger and bigger, until it reaches around 65535K (64
megabytes).  If you haven't altered your Perm Gen settings, then it will
start hanging.  Generally, this PermGen area never gets garbage
collected.

To alter your PermGen memory use the following VM argument (adjusting to
the megabyte limit you want):

 -XX:MaxPermSize=128m 

cheers,
Ants.

keith chapman wrote: 

Yes We had to increase the the PermGen memory for the JVM too on
http://mooshup.com which is an online version of the WSO2 Mashup Server
which is built on top of Axis2 essentially. We are running about 70
services (All of them are not aars, most of them are services written in
JavaScript which is our primary focus) there at the moment. And its
running on an EC2 box.

Thanks,
Keith.

On Tue, Apr 22, 2008 at 5:42 AM, Anthony Bull
<[EMAIL PROTECTED]> wrote:

I don't have the answers to this, but one thing to watch is your PermGen
memory for the JVM, we've had to increase this on our Tomcats to handle
lots of web services that have their jars included in the .aar (as
opposed to using common shared jars in WEB-INF/lib).  You get to a point
where your Tomcat or JVM just hangs permanently, often without error
messages. 



Dave MacLean wrote:

Hello,
We are using Axis2 to host a large number of web services contained in
.aar archives.  We have some code that will allow users to automatically
generate new services "on-the-fly" that leverages the hot-deployment
features in Axis2, and so we end up with potentially large numbers of
services deployed at a given time.

I was wondering if there have been any scalability tests done, or if
there is a guideline on approximately how much hardware is needed to
host X number of services, etc.  I have not been able to find anything
in the docs or mailing list archives.  Basically, I want to know when we
might start to run into performance problems, and how many services we
can realistically expect to host on a single server.

Is there any data available on this?  Anyone else running a similar
setup that can share their findings?

Thanks in advance,

Dave

---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]



 

 

-- 

 

--------------------------------------------------------------------- To
unsubscribe, e-mail: [EMAIL PROTECTED] For additional
commands, e-mail: [EMAIL PROTECTED] 

Reply via email to