[slurm-users] addressing NVIDIA MIG + non MIG devices in Slurm

Thu Jan 27 16:00:52 UTC 2022

Hi Mathias

I can't answer your specific question, so this is more of a comment 🙂

We have a system with 8 x Nvidia A40, where we would like to share each GPU between several jobs (they have 48GB each), eg starting 32 jobs, with 4 on each GPU. I looked into MIG as well, but unfortunately that is not supported by the A40 hardware (only A30 and A100).
I have tried MPS, but strangely that works only for the first GPU on each node, so only one of the 8 GPUs in our system can be shared in this way. That is, it used to be like that. A couple of weeks ago, an "all_sharing" flag was introduced for gres.conf, which apparently should make it possible to share all the GPUs with MPS. I haven't tried it yet, but it may be worth a try. It should be possible to configure some GPUs as mps and some as gpu resources.

Cheers,

Esben

https://github.com/SchedMD/slurm/blob/master/doc/man/man5/gres.conf.5

all_sharing
To be used on a shared gres. This is the opposite of one_sharing and can be
used to allow all sharing gres (gpu) on a node to be used for shared gres (mps).
NOTE: If a gres has this flag configured it is global, so all other nodes with
that gres will have this flag implied.  This flag is not combatible with
one_sharing for a specific gres.

________________________________
From: slurm-users <slurm-users-bounces at lists.schedmd.com> on behalf of Matthias Leopold <matthias.leopold at meduniwien.ac.at>
Sent: Thursday, January 27, 2022 16:27
To: Slurm User Community List <slurm-users at lists.schedmd.com>
Subject: [slurm-users] addressing NVIDIA MIG + non MIG devices in Slurm

Hi,

we have 2 DGX A100 systems which we would like to use with Slurm. We
want to use the MIG feature for _some_ of the GPUs. As I somehow
suspected I couldn't find a working setup for this in Slurm yet. I'll
describe the configuration variants I tried after creating the MIG
instances, it might be a longer read, please bear with me.

1. using slurm-mig-discovery for gres.conf
(https://eur02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgitlab.com%2Fnvidia%2Fhpc%2Fslurm-mig-discovery&data=04%7C01%7Cepf%40novozymes.com%7Ca0989487f14947bd269808d9e1a99cc4%7C43d5f49ee03a4d22a2285684196bb001%7C0%7C0%7C637788940862367005%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=AHN8h8%2FcB4xeC7MFdYgZRG7L62PTiz4OvampC1vWL5Q%3D&reserved=0)
- CUDA_VISIBLE_DEVICES: list of indices
-> seems to bring a working setup and full flexibility at first, but
when taking a closer look the selection of GPU devices is completely
unpredictable (output of nvidia-smi inside Slurm job)

2. using "AutoDetect=nvml" in gres.conf (Slurm docs)
- CUDA_VISIBLE_DEVICES: MIG format (see
https://eur02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdocs.nvidia.com%2Fcuda%2Fcuda-c-programming-guide%2Findex.html%23env-vars&data=04%7C01%7Cepf%40novozymes.com%7Ca0989487f14947bd269808d9e1a99cc4%7C43d5f49ee03a4d22a2285684196bb001%7C0%7C0%7C637788940862367005%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=ZFXeCQh3%2FLO52zpiOqn7CROjq7wvj8zWRE8mj87P%2Bew%3D&reserved=0)

2.1 converting ALL GPUs to MIG
- also a full A100 is converted to a 7g.40gb MIG instance
- gres.conf: "AutoDetect=nvml" only
- slurm.conf Node Def: naming all MIG types (read from slurmd debug log)
-> working setup
-> problem: IPC (MPI) between MIG instances not possible, this seems to
be a by-design limitation

2.2 converting SOME GPUs to MIG
- some A100 are NOT in MIG mode

2.2.1 using "AutoDetect=nvml" only (Variant 1)
- slurm.conf Node Def: Gres with and without type
-> problem: fatal: _foreach_slurm_conf: Some gpu GRES in slurm.conf have
a type while others do not (slurm_gres->gres_cnt_config (26) > tmp_count
(21))

2.2.2 using "AutoDetect=nvml" only (Variant 2)
- slurm.conf Node Def: only Gres without type (sum of MIG + non MIG)
-> problem: different GPU types can't be requested

2.2.3 using partial "AutoDetect=nvml"
- gres.conf: "AutoDetect=nvml" + hardcoding of non MIG GPUs
- slurm.conf Node Def: MIG + non MIG Gres types
-> produces a "perfect" config according to slurmd debug log
-> problem: the sanity-check mode of "AutoDetect=nvml" prevents
operation (?)
-> Reason=gres/gpu:1g.5gb count too low (0 < 21) [slurm at 2022-01-27T11:23:59]

2.2.4 using static gres.conf with NVML generated config
- using a gres.conf with NVML generated config where I can define the
type for non MIG GPU and also set the UniqueId for MIG instances would
be the perfect solution
- slurm.conf Node Def: MIG + non MIG Gres types
-> problem: it doesn't work
-> Parsing error at unrecognized key: UniqueId

Thanks for reading this far. Am I missing something? How can MIG and non
MIG devices be addressed in a cluster? This setup of having MIG and non
MIG devices can't be exotic, since having ONLY MIG devices has severe
disadvantages (see 2.1). Thanks again for any advice.

Matthias

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.schedmd.com/pipermail/slurm-users/attachments/20220127/ab4f4a31/attachment-0001.htm>