> Hi again Scott, > > >similar to yours, in that the average response time for a request > >remained fairly consistent. > > Does this mean that the amount of time to get a response back from a request was the > same in your test for 1 and mutliple transactions. I would of thought an a noisy > channel there would have been some performance issue?
The response time was not *exactly* maintained when adding more client threads, but it was closer to constant than what you describe. In my tests, network I/O was a very small part of the total round trip. Packet sizes were on the order of 1000 bytes. Over a 100 Mbps LAN, that's about 0.1 ms transport time, which is very small compared to the overall round trip times that were more like 90 ms. This may point out a difference in our tests: mine were relatively processing intensive, while yours may be more network intensive. > > >The configuration of your Web server, servlet container and even > >server OS can impact your results. All 3 likely have a pool of > >threads and/or memory structures from which resources are allocated > >to service requests. If the pools are small, your SOAP requests will > >sit in the queue longer. Making sure the pools are large enough is > >essential to obtain proper throughput. > > I'm running the soap server on a JRUN 4 server, do you have any advice on j2ee > settings. Do you have any advice on how to manipluate the pools and which pools! > What settings do usually amend on your j2ee app server, or elsewhere I don't have a copy of JRun 4 available. I used JRun 3 some time ago, but I understand 4 was a pretty significant re-write. With 3, however, I seem to recall that copies sold under development license were restricted in the number of worker threads. I don't know whether 4 is similar (or whether you are even using a development license). As I said, there may also be differences in our test services. Perhaps you are sending much bigger packets or doing much less processing. I would think that as the transport time and total round trip times get closer, you would start having increasing Ethernet collisions as you added client threads. Scott Nichol Do not reply directly to this e-mail address, as it is filtered to only receive e-mail from specific mailing lists.
