Unsolved
This post is more than 5 years old
1 Message
0
5416
November 21st, 2003 23:00
java does not scale on 2650 dual Xeon? Any tips appreciated
We're in the process of trying to decide whether we should add a cpu or not to our 2650s (2.4Gz Xeon processors)
We have a Poweredge 2650 with up to two 2.4GHz Xeon processors.
Details further below, but my questions are:
1) have you seen a non-trivial java server program scale well when going from 1 to more than one processor?
2) are there some obvious tools here that could help pinpoint the bottleneck? e.g., some perhaps obvious sanity checks we can do
Our own java application is basically a server. In performance testing, we did not see any improvement in performance when we added a 2nd cpu. Both cpu's were cranking (essentially fully utilized, as far as I could tell), but the net throughput in terms of requests processed did not.
A simple Java program that creates lots of threads that just loop (i.e., don't interact or share resources) WILL scale perfectly.
We came across a free Java benchmark that has been used by many other people (http://www.google.com/search?hl=en&ie=UTF-8&oe=UTF-8&q=volanomark) in the process of benchmarking jdk's and Linux - it's called VolanoMark (http://www.volano.com/benchmarks.html). It's a tester for a chat server, and hence stresses several things outside the realm of just cpu. However, it seemed to be a sufficiently similar thing to our own app, and so we're using it to see if we can at least see some scalability with that.
On the 2650 system, the VolanoMark tool (loopback test) does not scale with kernels RH 7.3 (latest), RH9, RH ES 3, or 2.6-test9. This is all WITHOUT hyperthreading, which seems to slow things down further.
Apparently this tool (VolanoMark) has scaled on smp machines (not PowerEdge, AFAIK). I even verified that it scaled on a friends smp box with the 2.6 kernel.
I realize there are lots of players here, besides just the processor. We're not getting any evidence that having a dual cpu will help, and so we'd end up getting extra servers. However, before making the ultimate decision I'm trying to get more feedback. Maybe we're missing something obvious to try - some "magical" setting somewhere - mebbe it could happen :)
We have a Poweredge 2650 with up to two 2.4GHz Xeon processors.
Details further below, but my questions are:
1) have you seen a non-trivial java server program scale well when going from 1 to more than one processor?
2) are there some obvious tools here that could help pinpoint the bottleneck? e.g., some perhaps obvious sanity checks we can do
Our own java application is basically a server. In performance testing, we did not see any improvement in performance when we added a 2nd cpu. Both cpu's were cranking (essentially fully utilized, as far as I could tell), but the net throughput in terms of requests processed did not.
A simple Java program that creates lots of threads that just loop (i.e., don't interact or share resources) WILL scale perfectly.
We came across a free Java benchmark that has been used by many other people (http://www.google.com/search?hl=en&ie=UTF-8&oe=UTF-8&q=volanomark) in the process of benchmarking jdk's and Linux - it's called VolanoMark (http://www.volano.com/benchmarks.html). It's a tester for a chat server, and hence stresses several things outside the realm of just cpu. However, it seemed to be a sufficiently similar thing to our own app, and so we're using it to see if we can at least see some scalability with that.
On the 2650 system, the VolanoMark tool (loopback test) does not scale with kernels RH 7.3 (latest), RH9, RH ES 3, or 2.6-test9. This is all WITHOUT hyperthreading, which seems to slow things down further.
Apparently this tool (VolanoMark) has scaled on smp machines (not PowerEdge, AFAIK). I even verified that it scaled on a friends smp box with the 2.6 kernel.
I realize there are lots of players here, besides just the processor. We're not getting any evidence that having a dual cpu will help, and so we'd end up getting extra servers. However, before making the ultimate decision I'm trying to get more feedback. Maybe we're missing something obvious to try - some "magical" setting somewhere - mebbe it could happen :)
No Events found!