JBossWS performance

[ters]

Hi,
I am trying to compare stateless beans' performance to JAX-WS web services. So, I deployed a very simple stateless bean with a simple "ping" method and used @WebService to expose it as a webservice on JBoss 4.0.5.

Using multiple (16) standalone clients from a client machine I can get around 1700 req/sec from the stateless bean deployed on a saturated 1GHz machine (95% cpu by JBoss, 18 Mbps in, 7Mbps out at peak).

Using multiple (6) axis clients from the client machine I can only get 190 req/sec under exact same settings on the 1GHz machine (97% cpu by JBoss, 1.7Mbps in, 1.6Mbps out at peak)

I know that Webservices are generally slow due to XML (de)marshalling but JAX-WS service is an order of magnitude slower than the stateless bean. Is my observation consistent with what other people are experiencing with JBoss?

Server: linux 2.6.17, java 1.5, JBoss 4.0.5

[ters]

"ters" wrote:
Hi,

Using multiple (6) axis clients from the client machine I can only get 190 req/sec under exact same settings on the 1GHz machine (97% cpu by JBoss, 1.7Mbps in, 1.6Mbps out at peak)

I realized that the axis client is using HTTP/1.0 requests which leads to TCP open-close for every request. I changed the axis http connector with jakarta commons connector. It didn't help with the req/sec but it brought network usage to 900Kbps in 1.2 mbps out on server.

[thomas.diesler]

Have a look at the JAX-WS docs preview

http://tddell.dyndns.org/mediawiki/index.php/Performance_Analysis