Preliminary Benchmark Results
A benchmark based on a chatroom simultor was used to evaluate the scalability of the priority queue and multi-queue scheduler implementations. This benchmark is similar to the VolanoMark  benchmark, except that the use of Java is totally eliminated.  The numbers reported on the y axis of the following graphs are average throuhgput, where bigger is better.  The numbers on the x axis indicate the number of CPUs in the system.

As a baseline, we first ran the benchmark on a standard 2.4.0-test10 kernel and received the following results.

Due to the large drop in throughput when going from 4 ot 8 CPUs, we decided to run the benchmark on a standard 2.4.0-test9 kernel.

As you can see the throughput numbers for 2.4.0-test10 at 4 CPUs were better for 2.4.0-test10 than they were for 2.4.0-test9.  However, when going from 4 to 8 CPUs the throughput numbers for 2.4.0-test10 dropped off much more dramatically than hey did on 2.4.0-test9.

Next, these benchmarks were run with the 2.4.0-test10 kernel with a priority queue scheduler impementation.

Finally, the tests were run on a 2.4.0-test10 kernel with a multi-queue scheduler implementation.

For ease of comparison, the results of the10 room/100 message run for all implementations are presented on the same graph.

Here the same comparison is made for a 30 room/300 message run.