| 29 Sep 2022 |
|  fzakaria joined the room. | 18:22:07 |
| fzakaria | Workshop on reproducibility
https://ucsc.zoom.us/j/92228537801?pwd=SGNmTHRaNjE3UURsRjNHQzZGektRZz09
Title: Reproducibility in HPCThere is a reproducibility crisis across the scientific research world — studies have shown that the results of many scientific papers cannot be reproduced. In HPC the problem is exacerbated, as each machine is essentially unique. Performance results depend on the CPU, GPU, and system software and hardware configuration. It is seldom the case that we run the exact same configuration of some program twice.Are there solutions that can enable better reproducibility for HPC, or are we looking at the problem the wrong way? Do we care about reproducing a particular build, or a particular conclusion? This panel will explore issues of software and hardware reproducibility in HPC.Our 4 panelists will present some different angles on reproducibility, and we’ll follow this up with some questions and discussion:
| 18:22:25 |
 tpw_rules | is there some sort of talk announcement for this? i joined but just want some context | 19:25:35 |
| fzakaria | last minute -- sorry | 19:26:42 |
| fzakaria | it was : https://ospo.ucsc.edu/post/20220823/ | 19:26:47 |
| fzakaria | or
https://docs.google.com/document/u/1/d/e/2PACX-1vQin8rKcMZhWwi31e2Djbbvgo8uAWexyRixqQHUgEpmeXGn5As465yKhZl8IEpvuBmDfzly9ybxo5w8/pub | 19:27:15 |
| fzakaria | we are on the tail end now | 19:27:33 |
| tpw_rules | ah, you are presenting? | 19:27:51 |
| tpw_rules | i have used nix on hpc before but more just as a curiosity | 19:28:53 |
| tpw_rules | but demonstrating that it's practical to use without administrator collaboration | 19:31:56 |
| tpw_rules | oh that's what you meant by tail end | 19:34:26 |
| tpw_rules | how did it go? is there a recording? | 19:34:33 |
| tpw_rules | * oh that's what you meant by tail end. i misread your message's time. | 19:34:56 |
|  @rimuru:gentoo.chat changed their profile picture. | 21:30:33 |
| 30 Sep 2022 |
|  bryan changed their profile picture. | 12:16:42 |
|  louib joined the room. | 13:04:32 |
| fzakaria | i think there will be a recording published + slides | 16:04:46 |
| fzakaria | the part by Todd Gamblin on their use of sat solving for version picking is very cool | 16:05:01 |
| fzakaria | in Spack they allow different versions of the dag, and they use sat solving to come up with correct builds or even minimal rebuilds according to what's in the cache already | 16:05:24 |
| fzakaria | https://spack.io | 16:05:31 |
| tpw_rules | how does it constrain things? version numbers? | 16:55:30 |
| tpw_rules | * how does it constrain things? version numbers? i assume that's what "version picking" means | 16:58:55 |
 tnias | you can provide constrains like what packages you are interested in and (optionally) their version numbers, compiler flags, etc.
then there is a sat solver (they call it concretizer) that provides a solution to the constrains. This solution is saved as a *.lock file. From a lock file you can reproduce the environment.
packages can also provide their own constraints. as in: "version x of this package needs gcc >= 9"
| 17:29:27 |
| tnias | (at least that is my understanding of the talk) | 17:30:56 |
|  @0b11:matrix.org removed their display name 0x53. | 17:33:15 |
| @0b11:matrix.org left the room. | 17:33:26 |
 afontain | how is that different from what debian does ? | 17:38:18 |
| afontain | someone had fun solving sudoku puzzle with debian packages constraints | 17:38:51 |
|  0xMRTT [envs.net] joined the room. | 18:43:32 |
| fzakaria | let me share his slide deck | 22:00:13 |
