BOINC (WCG) performance across distros

[lazlo_vii]

I have read some of the articles on phoronix.com where various Linux distros are benchmarked to compare their performance. I have a few systems running various Linux distros along with BOINC and World Community Grid Project. Because the executable files running WCG projects are out outside of the control of the distro maintainers and any benchmarking suite they are essentially "Black Box" workloads. I am wondering if anyone here has done testing to see which modern Linux distro gives the best bare metal performance (measured in results returned per day) for World Community Grid tasks.

Thanks for reading, and enjoy your day!

[Michael]

Is there any way to run BOINC WCG as an automated benchmark test? If so details on how to do so for making a test profile in future distro comparison?

[lazlo_vii]

To my knowledge, no. When it is first installed the BOINC Client will run a Whetstone benchmark on the system to get a rough estimate of single core performance. Other than that there are no automated benchmarks in BIONC that I know of.

[lazlo_vii]

Some more information about the WGC tasks might be handy. First off, they all run as single threads on the system at the lowest priority level but using 100% execution time. Next, they are designed to run on a very broad spectrum of hardware so they don't take a lot of RAM. Some people claim SMT helps get more work done and others claim it does not. The tasks cover a wide variety of scientific research with most of them focused on biology, medicine, and ecology. Because of this the time it takes to run work units within a Sub-Project is fairly consistent but the run time can vary widely between Sub-Projects. It would likely be best if any tests were done over over a period of a few days or a week before switching distros in order to account for these variations.

[lazlo_vii]

I have done some testing on my own and even though it hasn't been as thorough as I would like, I think I have valid results. So far I have tested Ubuntu (18.04 and 20.04), Devuan, and Clear Linux on an Intel Xeon D1520. On all four versions the results are so similar I cannot detect any difference what so ever. If I had a more powerful system with a higher core count, higher clock speed, and allowed longer runs some difference might appear. As it is, I want to spend my CPU cycles crunch data so I will stop here.