[NLPL Task Force (A)] Saga usage

[Artur Kulmizev]

Hi Stephan,

I’m sorry to push back, but here’s a bit of clarification (if it helps): I only use Saga for a single project, only because it’s less restrictive with virtual environments than Puhti. With the exception of the jobs I have running at this current moment (19 parsing jobs, which should each finish in <6 hours), the only jobs I run on the cluster are simple embedding loaders, which all take about 12 minutes to run on GPU. Maybe I don’t understand the slurm queuing system exactly, but what would stop someone’s job from running after mine are (promptly) finished? For contrast, I remember several users running many 48-hour+ jobs this past spring, which effectively locked me out of the system for days at a time. Also, I haven’t had much luck on the CPU nodes, which often have tighter queues than the GPU ones.

Again, I don’t mean to be combative about this. I guess I just don’t think that my usage puts a heavy load on the system, when compared with other users (despite what the queue might indicate). In any case, I’m not a local (Norwegian) user, and if this is a big problem for others that are local, I can move to a different cluster if necessary.

Best,

-Artur

---
Artur Kulmizev
PhD Student, Computational Linguistics Group
Uppsala University
Dept. of Linguistics and Philology
Box 635
SE-751 26 Uppsala
Sweden



> On 21 Aug 2020, at 15:05, Stephan Oepen <oe at ifi.uio.no> wrote:
>
> dear both:
>
> it appears you are making good usage of Saga, and that is what we
> intend through this nordic partnership.  however, Saga has been very
> heavily loaded this summer, and the national administrators are asking
> all projects to try and reduce usage, for example cutting out strictly
> speaking unnecessary 'bonus' jobs or spreading out their computing
> over longer periods of time.
>
> could i kindly ask you to try and scale back the load you put on Saga
> a little, in particular aim to not line up too many jobs in the gpu
> queue, which practically means other users may find themselves 'locked
> out' for extensive wait periods.
>
> arradi: as an MSc student, i would ask that you typically limit your
> concurrent gpu usage to at most five active jobs (or, correspondingly,
> a maximum of five gpus).  just now, that would mean that you should
> terminate or cancel 15 jobs and wait to resubmit these until either
> your other jobs have completed, or the gpu queue becomes partly idle.
>
> i am sorry to have to impose such 'soft' usage constraints, but
> recently the situation on Saga has been unsatisfactory to some users,
> and we need to avoid any impression of individual user communities
> getting into the way of others.
>
> with best wishes, oe









När du har kontakt med oss på Uppsala universitet med e-post så innebär det att vi behandlar dina personuppgifter. För att läsa mer om hur vi gör det kan du läsa här: http://www.uu.se/om-uu/dataskydd-personuppgifter/

E-mailing Uppsala University means that we will process your personal data. For more information on how this is performed, please read here: http://www.uu.se/en/about-uu/data-protection-policy