[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Any input for some talk about usage of Debian in HPC



Hi Andreas,

On 2024-05-19 05:12, Andreas Tille wrote:
While I could talk about Debian Science and Debian Med in general it
would be cool to reference to some real life examples where Debian is
used in Science and what might be the reason to use Debian.

I personally would like to stress the "we package what we use" aspect
and the "we mentor upstream to merge competence of the program with
packaging skills" idea.  Any input would be welcome to cover more ideas.

It is perhaps not quite what you're asking for, but Debian has had a positive effect on the ROCm compute stack used on AMD GPU-based systems such as the Frontier and LUMI supercomputers. These systems are not running on Debian, but their compute stack has nevertheless benefited from feedback provided by Debian packagers.

The high standards of the Debian project have helped to identify numerous issues that were resolved upstream to the benefit of users of all Linux distributions. For example, the symbol tracking done by Debian packagers revealed that there were many weak symbols in the ROCm math libraries that were not being hidden as intended. Library authors did not expect the symbols for GPU kernels to be visible, so they were therefore not being namespaced, which made it a problem when the symbols appeared in the public ABI. This was fixed upstream due to feedback from Debian packagers.

Another example would be the copyright review for the rocBLAS library, which discovered a number of kernels that did not include source code. While the library was intended by upstream to be fully open-source, the authors used a few kernels written in a proprietary shader language and built with a proprietary toolchain. The disassembles of these kernels were then released under the Expat license. To the authors, this seemed reasonable, as the library contained many hand-written assembly kernels anyway. While other operating systems had packaged rocBLAS before Debian, it was the Debian reviewers who flagged these kernels as violating the DFSG and brought the issue to upstream. The closed-source kernels were removed from the library upstream after enhancing the Tensile kernel generator to prevent performance regressions.

On a tangential note, I would like to mention that the Oregon Advanced Computing Institute for Science and Society (OACISS) is currently hosting an MI210 GPU server [1] that AMD provided with the explicit purpose of facilitating the packaging and testing of GPU-accelerated software on Debian. I should probably have been more organized about preparing an official announcement, but I hope this email will help to get the word out. If there are any Debian Developers that require access to AMD GPU hardware to enhance software on Debian, please let me know and I can help. AMD has provided a variety of hardware in the past, and the OACISS has been generous in providing access to HPC hardware for the Debian community.

Sincerely,
Cory Bloor

[1]: https://salsa.debian.org/rocm-team/community/team-project/-/wikis/Continuous-integration-workers#pinwheel


Reply to: