Skip to main content

Cluster optimizations?

7 replies [Last post]
samdoyle
Offline
Joined: 2006-12-27

My deployed cluster is running incredibly slow to the point which it is horrible. The standalone glassfish instance which also acts as the web service has to be running at least 10x faster then then the loadbalanced/clustered deployment.

Is there any optimization settings that I might be missing?

Reply viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.
samdoyle
Offline
Joined: 2006-12-27

I'm also seeing all requests being directed to only one instance in a two instance cluster with round-robin policy.

The load balancing is performing absolutely horribly on a quad core processor running 64bit Fedora 8 and this is with only a few minor requests.

samdoyle
Offline
Joined: 2006-12-27

It looks as if the availability option makes a huge impact on the performance. As I mentioned only a few requests resulted in a user experience that was pretty much unusable. When I disable the availability it is now much better but still significantly slower then a non-clustered environment.

S.D.

Gabor Szokoli

Hi,

Just interested:

On Nov 26, 2007 9:13 AM, wrote:
> It looks as if the availability option makes a huge impact on the performance. As I mentioned only a few requests resulted in a user experience that was pretty much unusable. When I disable the availability it is now much better but still significantly slower then a non-clustered environment.

Do you see heavy load (CPU or IO) on your servers, or just slow responses?

Szocske

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@glassfish.dev.java.net
For additional commands, e-mail: users-help@glassfish.dev.java.net

samdoyle
Offline
Joined: 2006-12-27

Let me look.

Thanks, S.D.

samdoyle
Offline
Joined: 2006-12-27

When logging in/initializing the application with availability enabled two of the cores encountered a substantial spike. One went to 74% the other 62% but that is about all I saw. Both instances, for purposes of testing, are running on the same server.

S.D.

samdoyle
Offline
Joined: 2006-12-27

I just ran iostat as well and noticed a spike of 82 I/O requests/second.

After disabling the availability the number of I/O requests appeared to be close but the CPU load was substantially lower. The two cores I mentioned before went to 28% and 11% load.

S.D.

samdoyle
Offline
Joined: 2006-12-27

Ah but then again I just did a CTRL-F5 on the login page with availability enabled. There was no real spike on the server but yet the page took 15 seconds to load so I'm not really sure so in that case it was just dog slow response time.

S.D.