为多个同时的SOAP请求配置Tomcat

Question

I'm very much a Tomcat newbie, so I'm guessing that the answer to this is pretty straightforward, but Google is not being friendly to me today.

I have a Java web application mounted on Apache Tomcat. Whilst the application has a front page (for diagnostic purposes), the application is really all about a SOAP interface. No client will ever need to look up the server's web page. The clients send SOAP requests to the server, which parses the requests and then looks up results in a database. The results are then passed back to the clients, again over SOAP.

In its default configuration, Tomcat appears to queue requests. My experiment consisted of installing the client on two separate machines pointing at the same server and running a search at exactly the same time (well, one was 0.11 seconds after the other, but you get the picture).

How do I configure the number of concurrent request threads? My ideal configuration would be to have X request threads, each of which recycles itself (ie calls destructor and constructor and recycles its memory allocation) every Y minutes, or after Z requests, whichever is the sooner. I'm told that one can configure IIS to do this (although I also have no experience with IIS), but how would you do this with Tomcat?

I'd like to be able to recycle threads because Tomcat seems to be grabbing memory when a request comes in and not releasing it, which means that I get occasional (but not consistent) Java Heap Space errors when we are approaching the memory limit (which I have already configured to be 1GB on a 2GB server). I'm not 100% sure if this is due to a memory leak in my application, or just that the tools that I'm using use a lot of memory.

Any advice would be gratefully appreciated.

Thanks, Rik

Answer 1

Tomcat, by default, can handle up to 150 concurrent HTTP requests - this is totally configurable and obviously varies depending on your server spec and application.

However, if your app has to handle 'bursts' of connections, I'd recommend looking into Tomcat's min and max "spare" threads. These are threads actively waiting for a connection. If there aren't enough waiting threads, Tomcat has to allocate more (which incurs a slight overhead), so you might see a delay.

Also, have a look at my answer to this question which covers how to configure the connector:

Tomcat HTTP Connector Threads

In addition, look at basic JVM tuning - especially in relation to heap allocation overhead and GC pause times.