Hacker News Reader

Comment by VeninVidiaVicii | Hacker News Reader

As a computational biologist, I've been struggling with performance issues on Ubuntu for months, which have made it difficult to work efficiently. Despite Ubuntu's popularity in academia, R runs much slower on Ubuntu than on my personal computer running Arch. This has forced me to do my work at home, where I have better performance. It's disappointing that Ubuntu's limitations are often assumed to be universal to Linux in academia, which can be a problem for those who require a more efficient system. I've tried troubleshooting the issue, but haven't found a solution yet, but just chock it up to Ubuntu just being more and more terrible

View on HN

> I've tried troubleshooting the issue, but haven't found a solution yet

That is very strange. I assume you’re not using the same hardware in your home computer. Without knowing your setup, it’s hard to say anything. Are you using the same BLAS and LAPACK libraries? You can check with `sessionInfo()`.

That’s a very interesting thought, and one I haven’t come across. How did you come up with this so fast?

I read about it before.[1] There is also a BLAS library made by Nvidia that makes use of CUDA. This post is about how to hook it up to GNU Octave but it also works with R:

https://developer.nvidia.com/blog/drop-in-acceleration-gnu-o...

What sibling says about compiling R on your work computer so that it can make full use of your CPU can help too. I am curious if that closes the performance gap for you.

[1] https://csantill.github.io/RPerformanceWBLAS/

I’ll give it a shot when I make it back to campus. I’m snowed in at home, but I’ll let you know soon. Also, this really does look like the issue. Thanks for the insightful comments.

These tips speed up my Arch Linux R even more, but my Ubuntu setup is still just as slow, even with these optimizations. Somewhere along the journey I did see that Xeons have problems with R, but I can't seem to find where I saw that. Darn.

Is it maybe that the default R binary on Ubuntu hasn't been optimized? Have you tried building R from source on Ubuntu?

I've personally tried building the software I need for source before, but getting all the BLAS/LAPACK and MPI libraries with it has been annoying and time consuming. Your supercomputer cluster may offer pre-built optimized binaries for your CPU. For mine, I've been using prebuilt software offered by compute canada over CVMFS[1]. I assume something similar is available for wherever you get your supercomputing resources from.

1. https://docs.alliancecan.ca/wiki/Accessing_CVMFS

> I've been struggling with performance issues on Ubuntu for months

I've run Ubuntu on various hardware and in VM's since 2006 running a wide variety of applications, there are no performance issues I've seen. Perhaps R is different to all other software I've run (which is a lot).

Reply on news.ycombinator.com