Re: [HTCondor-users] Question about VirtualMemory

[Di Qing <dqing@xxxxxxxxx>]

Hi All,

Any clues on this? It sounds like a bug since the behavior (setting 
memory.memsw.limit_in_bytes with cgroup) is inconsistent between 
partitionable and non-partitionable slot types. For us, ideally the slot 
should be a partitoionable slot with some preassigned physical memory 
and swap space, and the job running there should receive the same swap 
space percentage according to the physical memory it requests. However, 
"PROPORTIONAL_SWAP_ASSIGNMENT=true" doesn't work, and in my test, the 
memory.memsw.limit_in_bytes was always set to total virtual memory of 
the whole node when the slot type is partitionable as I described in my 
previous email.

By the way, I tested setting VirtualMemory with SUBMIT_EXPRS as:

SUBMIT_EXPRS = $(SUBMIT_EXPRS) VirtualMemory
VirtualMemory = RequestMemory

it did  not work either and it seems that HTCondor just ignores 
VirtualMemory in job JDL.

Thanks,

Di Qing

On 03/02/17 03:07 PM, Di Qing wrote:
> Hi All,
>
> I am trying to find a way to limit the memory usage of the jobs and
> avoid they use too much swap space. I am testing HTCondor 8.4.11 with
> CGroup under SL6. The node has been partitioned to two static slots with
> different memory and CPU assignment. Each static slot can be further
> partitioned into smaller dynamic slots according to the CPU and memory
> requests of jobs.
>
> In my test, with hard limit of CGroup, when configuring the slot
> partitionable, when job running, the memory.limit_in_bytes is set to the
> memory requested by the job and memory.memsw.limit_in_bytes is set to
> the total virtual memory of the node, i.e., physical memory plus swap
> space, although I configured its swap space to some percentages or set
> PROPORTIONAL_SWAP_ASSIGNMENT to true. If I configured the slot is not
> partitionable, the memory.limit_in_bytes of job is set to the assigned
> physical memory of
> slot and memory.memsw.limit_in_bytes is set to the virtual memory of the
> slot.
>
> With soft limit of CGroup,  both memory.limit_in_bytes and
> memory.memsw.limit_in_bytes set to the total virtual memory of the node
> (not the slot) when configuring the slot partitionable. Also the
> memory.soft_limit_in_bytes is set to the memory requested by job. When
> the slot is not partitionable, both memory.limit_in_bytes and
> memory.memsw.limit_in_bytes set to the virtual
> memory assigned to the slot, and memory.soft_limit_in_bytes is set to the
> physical memory of slots.
>
> My question is why HTCondor always sets the  memory.memsw.limit_in_bytes
> to the total virtual memory of the node not the slot when the slot is
> configured partitionable? We don't want to completely disable the swap
> device since OS or some light tasks can still use some swap space.
>
> Thanks,
>
> Di
> _______________________________________________
> HTCondor-users mailing list
> To unsubscribe, send a message to htcondor-users-request@xxxxxxxxxxx with a
> subject: Unsubscribe
> You can also unsubscribe by visiting
> https://lists.cs.wisc.edu/mailman/listinfo/htcondor-users
>
> The archives can be found at:
> https://lists.cs.wisc.edu/archive/htcondor-users/