[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [HTCondor-users] Problems Defining Additional Slot Types

I am using NUM_CPUS I just accidently left off the "S" when typing the email. However, I was leaving off the plural of SLOTS when trying the NUM_SLOTS_TYPE_1. Instead of $(MEMORY) which isn't defined I had to use $(DETECTED_MEMORY).

So, I made those changes and restarted and saw the change. Next, however, I was trying something like this


SLOT_TYPE_1 = cpu=1,ram=auto

SLOT_TYPE_2 = cpu=1,ram=16384

I was thinking this would give me the versatility having either 20 small memory slots, or ~12 large memory slots or a mixture depending upon what was submitted. (See my trailing question on my original post). However, this configure was not liked. It appears to make the 12 large memory slots but then it can't make the Type 1 slots.  >95% of the jobs that will run only need the smaller memory footprint, I don't have to have to sacrifice 8 cores most of the time just for the rare ability to need to run larger memory cases. I was under the impression that Condor would track the cpu/memory requests of jobs that it submits to slots and then not allow any more to run when it would exceed the supposed limits of either the number of cores or memory assuming full usage for each consumed slot.

Likewise, I would like to eventually create slots that are available for multiple-cpu (core) cases. For example, say I would like to have 2 slots available each with 10 cores and 1/2 the system memory. I certainly don't want only 2 slots available when a large number of single core jobs are submitted. This makes it seem that I can't fully utilize the versatility of the machine based on the potential job types that could be dynamically submitted.



Date: Mon, 4 Apr 2022 16:40:35 +0000
From: John M Knoeller <johnkn@xxxxxxxxxxx>
To: HTCondor-Users Mail List <htcondor-users@xxxxxxxxxxx>
Subject: Re: [HTCondor-users] Problems Defining Additional Slot Types
Content-Type: text/plain; charset="us-ascii"

When you define NUM_SLOTS_TYPE_1, the default type 0 slot will be disabled, so you would still only have one slot type.   I would expect that with a config of

SLOT_TYPE_1 = ram=16384

you would end up with 12 slots, each having 16384 Mb of memory.

If you don't configure NUM_SLOTS_TYPE_1,  then SL0T_TYPE_1 will be ignored.
Note I used NUM_SLOTS_TYPE_1 , not NUM_SLOT_TYPE_1.   plural, not singular.

Similarly, the knob for setting the number of cpus is NUM_CPUS,  not NUM_CPU.