Extend N-copies to 2N for Context Switching case in out-of-box run

[pdeng6]

This change proposes to extend the N-copies to 2N-copies for context switching case in out-of-box run(“copies” is not explicitly specified).

Currently, the out-of-box benchmark run(via ‘./Run’, where ‘-c’ option is not used) launches 1 copy and N-copies test, where N is the number of CPUs.

We investigated the performance characteristic on multiple high CPU core count systems and found that N-copies test of Pipe-based Context Switching case doesn’t saturate the system, e.g. in a 160 cores system, it is ~69%, and in a 224 cores system, it is ~62%. In this situation, the N-copies test doesn’t demonstrate the real system capability.

To address this problem, in this pull request, we propose to run a 2N-copies Pipe-based Context Switching case when ‘-c’ option is not used, others keep N-copies, and when ‘-c’ option is used, run the copies specified.

The 2N is an experience number that obtained from our tests in 7 systems where core number ranges from 96 to 288, with Linux kernel version 5.x and 6.x. According to the tests:

1. All systems are saturated with 2N copies.
2. Actually, all systems are saturated with copies less than 2N, say X. From X to 2N even to 4N, the score of context switching almost keeps flat with very slight drop which is less than 5%.

Low core count systems with 4/8/16 cores were also evaluated, 2N-copies saturated all, with less than 9.4% drops compare with N-copies.

I don’t have a good way to find the best X without running many times in a system, while in my humble opinion, 2N is a reasonable number since it almost demonstrates the real system capability and fair for different systems.

Please take a look, and very glad to hear your thoughts, thanks!