Re: [HTCondor-users] lower latency scheduling

[Dudu Handelman <duduhandelman@xxxxxxxxxxx>]

Rita. 

I think that 60 seconds for is a lot for 30 executions points. 

My Negotiation cycle is much lower for 1k ep.

Do you have many submitters or many autoclusters? 

I promise you that all the information you need is in the negotiator log

I would suggest to verify you have configured the knobs and make sure that you are on fast cpu with multiple cores. 

NEGOTIATOR_NUM_THREADS

NEGOTIATOR_CYCLE_DELAY

David 

Get Outlook for Android

---

From: HTCondor-users <htcondor-users-bounces@xxxxxxxxxxx> on behalf of Rita <rmorgan466@xxxxxxxxx>
Sent: Sunday, May 28, 2023 7:19:43 PM
To: HTCondor-Users Mail List <htcondor-users@xxxxxxxxxxx>
Subject: Re: [HTCondor-users] lower latency scheduling

 

Hi,

yes, I am using partitionable slots. I believe the current negotiation cycle is 60 secs. 

On Fri, May 26, 2023 at 10:52âAM Dudu Handelman <duduhandelman@xxxxxxxxxxx> wrote:

Rita. 

Are you using pslots? 

What is the current negotiation cycle time? 

It will show if the negotiation cycle is reasonable. 

Thanks 

David 

Get Outlook for Android

---

From: HTCondor-users <htcondor-users-bounces@xxxxxxxxxxx> on behalf of Rita <rmorgan466@xxxxxxxxx>
Sent: Friday, May 26, 2023, 12:20
To: HTCondor-Users Mail List <htcondor-users@xxxxxxxxxxx>
Subject: Re: [HTCondor-users] lower latency scheduling

CLAIM_WORKLIFE is a confusing term. By default its -1, "infinite claim worklife". I suppose I can put this as 10 (?). 

 I have already decreased SCHEDD_INTERVAL to 10 secs and I will do it for NEGOTIATOR_INTERVAL. Does the scheduler/negotiator throttle how many jobs can be scheduled per negotiation?

On Thu, May 25, 2023 at 5:38âPM Cole Bollig via HTCondor-users <htcondor-users@xxxxxxxxxxx> wrote:

Hi Rita,

Once a Schedd has a claim on Startd resources for a user the Schedd will try to run as many jobs that it can on those resources up to CLAIM_WORKLIFE seconds. Increasing the CLAIM_WORKLIFE would cause those claims to stay alive longer and get user jobs running once the resources are available again.

If you truly want to just increase the amount of negotiation cycles in the hopes of making more matches to resources, then you could try reducing the SCHEDD_INTERVAL and NEGOTIATOR_INERVAL.

-Cole Bollig

---

From: HTCondor-users <htcondor-users-bounces@xxxxxxxxxxx> on behalf of Rita <rmorgan466@xxxxxxxxx>
Sent: Thursday, May 25, 2023 5:29 AM
To: HTCondor-Users Mail List <htcondor-users@xxxxxxxxxxx>
Subject: [HTCondor-users] lower latency scheduling

 

I am trying to think of ways to lower my scheduling latency. My machines on Azure are relatively powerful. 256 cores, 2TB of memory, fast NVME local disks.  I have 30 of them in my HTcondor pool. 

My job profile is: they are small running jobs. each take like 30s to 5 mins to run. When the user submits the job, I want the scheduler/negotiator to aggressively match. No preemption. What are  some knobs we can tune for a setup like this? 

--

--- Get your facts first, then you can distort them as you please.--

_______________________________________________
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/

--

--- Get your facts first, then you can distort them as you please.--

_______________________________________________
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/

--

--- Get your facts first, then you can distort them as you please.--