[Beowulf] [External] LSF vs Slurm

[Prentice Bisbal]

Lohit,

I've been using Slurm for about 5 years now. I've used Sun Grid Engine 
and LoadLeveler in the past. At one site, they  were using Torque  
before I switched us to SGE, and at my current job they were using 
Torque/Maui before I found myself responsible for rolling out Slurm 
shortly after getting hired.

Based on that experience, here's my opinion of Slurm: It's much better 
than the others I've worked with. All the others I worked with had some 
kind of functional deficiency somewhere, that had to be fixed with 
various workarounds. For example, at the time I used SGE, it didn't 
support requesting exclusive use of a node, but there was a trick to 
make it do that. Related, most of these older schedulers seemed to have 
been created before MPI was really a thing, and you had to sometimes do 
some extra work for MPI jobs to run (use wrapper scripts, etc.).

I don't see any of those problems with Slurm. It seems like it was 
created from the start for modern HPC jobs and workloads. Just about 
every feature you would need for modern HPC is built right into it - no 
need for workarounds/kludges. And on top of that, it's fast! With SGE, I 
remember it would sometimes be a minute or two before a job started, but 
it seems like in most instances, when I submit a job to slurm it starts 
running immediately if the necessary resources are available. Now that 
could just be my luck with submitting jobs at the start of a scheduler 
run, but I know of other Slurm admins who have commented on how fast it is.

Finally, Slurm seems like it's become the 900-lb gorilla of the 
scheduler world, at least in academic and government research where I 
work, so there's a huge user base and user community to benefit from.

Our motivation was two-fold: savings on licensing, and using a newer 
scheduler that our collaborators at other sites and main campus were 
using. We do pay for support, but it's been used very rarely - I would 
have no problem supporting Slurm without it.

Prentice

On 3/10/22 11:39 AM, Lohit Valleru via Beowulf wrote:
> Hello Everyone,
>
> I wanted to ask if there is anyone who could explain me the benefits 
> of moving to Slurm from LSF.
>
> If anyone moved to Slurm, I want to ask on what 
> requirements/challengers motivated the move to Slurm, apart from 
> license cost with LSF.
>
> Thank you,
> Lohit
>
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin Computing
> To change your subscription (digest mode or unsubscribe) visit https://beowulf.org/cgi-bin/mailman/listinfo/beowulf