Computational performance RStudio Server slower than expected

[HeiN3r]

Hey everyone,

I have a rather general question regarding computational performance of the RStudio Server (open source edition).

Although being just an experiment at the moment, our IT department installed the server on a quite potent machine, which should in theory outrun any individual PC or laptop usually available at our offices. Despite this fact, the server computes and executes scripts slower than an ordinary windows PC. We benchmarked this using the benchmarkme package.

My server admin looked into this and found that on the Windows machine more cores seem to be utilized while on the Linux RStudio-Server only one core seems to be used (not using any explicit parallel coding). This would explain the differences.

I am a bit lost here. Is there a way to configure the server so that it uses all the potential computational power (similar to what the windows machine seems to be doing)? I really want to keep my RStudio-Server, but this hinges on the server delivering a performance boost (and not an actual loss).

Any insight would be greatly appreciated.

Thanks in advance.

[andrie]

What version of R do you have installed on Windows? Specifically, are you using the Intel MKL for parallel BLAS operations?

You can also enable parallel BLAS on Linux machines. You can achieve this by:

• Compiling R from source and activate a BLAS such as openBLAS or Atlas
• Install R from pre-compiled binaries (including) for Linux. These binaries are pre-compiled with OpenBLAS

[HeiN3r]

Thanks for the quick reply.
I am not "actively aware" about using parallel BLAS on the windows machine. I just installed a plain old 3.6.3 using the windows installer, nothing extra. If parallel BLAS is already integrated there, then yes. 3.6.3 is also the R version on the server.

But in any case, it would seem like a good idea to activate parallel BLAS for the Linux machine.

[HeiN3r]

As it turned out, the R version on the RStudio-Server was already compiled with BLAS. This would rule out this explanation of weak performance, I guess.