1.5 to 2.0 transactions a second? Try XFire
October 31st, 2005Lots of people wonder if SOA is a step back soley because of the performance. When I see statements like these I wonder too though:
The problem, Cohen said, is that Web services built with today’s commercial and open source J2EE application servers get 1.5 to 2.0 transactions per second (TPS), according to his tests, which he said is not good enough for production demands.
The good news, there is hope. I did some rather unofficial benchmarks of XFire last week. Now, I’m not going to post the full results here simply because I don’t have time to clean things up and hate benchmark debates. But here are a couple of the highlights:
- 25-85 messages a second
- 2-6 times as faster as Axis 1.3
- 1/2-1/5 the latency of Axis 1.3
Those benchmarks are over messages sizes from 8K to 80K. Once you get into the megabytes the differences will be even more dramatic. Oh and these are on my crappy Dell box which has a slow front side bus. My new laptop typically goes about 50% faster.
November 3rd, 2005 at 11:55 am
[...] I ran across this entry on XFire performance from Tim Pokorny by accident yesterday. While I feel my numbers speak for themselves, its always good to hear it from a user: …I decided to run only one client for 60 seconds and only on the AMD. It was able to get 3919 iterations completed in those 60 seconds. That is more than 65 updates per second. Perhaps the most amazing thing about this is the actual performance of web services (XFire in this case). These were not small updates either; two entities, each with something like eight attributes were being serialised into XML, sent over the wire and de-serialised into Java objects 65 times per second! That is HUGE. To put this in perspective, when previously asked by people about the real performance of web services (in the simulation context) I estimated that I could consistently get about 8 updates a second. Needless to say, 65 is almost an order of magnitude faster than that. [...]