[jira] [Comment Edited] (YARN-1051) YARN Admission Control/Planner: enhancing the resource allocation model with time.

2017-12-22 Thread Subru Krishnan (JIRA)

[ 
https://issues.apache.org/jira/browse/YARN-1051?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16302033#comment-16302033
 ] 

Subru Krishnan edited comment on YARN-1051 at 12/22/17 10:42 PM:
-

[~xingbao], the behavior depends on whether there's any job that's using more 
than it's guaranteed resources in the specific node and if preemption is 
enabled or not in the cluster. 

If there's no job using excess resources in the specific node, then either:
* relax locality to rack
* wait for one of the running job AMs to release container(s)

If there is at least one job which is using excess resources in the specific 
node, then:
* If you have preemption is enabled (refer [here | 
http://hadoop.apache.org/docs/current/hadoop-yarn/hadoop-yarn-site/CapacityScheduler.html#Capacity_Scheduler_container_preemption]
 on how to enable it), the over allocated container(s) will get preempted
*  wait for one of the running job AMs to release container(s)


was (Author: subru):
[~xingbao], the behavior depends on whether there's any job that's using more 
than it's guaranteed resources in the specific node and if preemption is 
enabled or not in the cluster. 

If there's no job using excess resources in the specific node, then either:
* relax locality to rack
* wait for one of the running job AMs to release container(s)

If there is at least one job which is using excess resources in the specific 
node, then:
* If you have preemption is enabled (refer 
[http://hadoop.apache.org/docs/current/hadoop-yarn/hadoop-yarn-site/CapacityScheduler.html#Capacity_Scheduler_container_preemption|here]
 on how to enable it), the over allocated container(s) will get preempted
*  wait for one of the running job AMs to release container(s)

> YARN Admission Control/Planner: enhancing the resource allocation model with 
> time.
> --
>
> Key: YARN-1051
> URL: https://issues.apache.org/jira/browse/YARN-1051
> Project: Hadoop YARN
>  Issue Type: Improvement
>  Components: capacityscheduler, resourcemanager, scheduler
>Reporter: Carlo Curino
>Assignee: Carlo Curino
> Fix For: 2.6.0
>
> Attachments: YARN-1051-design.pdf, YARN-1051.1.patch, 
> YARN-1051.patch, curino_MSR-TR-2013-108.pdf, socc14-paper15.pdf, 
> techreport.pdf
>
>
> In this umbrella JIRA we propose to extend the YARN RM to handle time 
> explicitly, allowing users to "reserve" capacity over time. This is an 
> important step towards SLAs, long-running services, workflows, and helps for 
> gang scheduling.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

-
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org



[jira] [Comment Edited] (YARN-1051) YARN Admission Control/Planner: enhancing the resource allocation model with time.

2014-03-20 Thread Arun C Murthy (JIRA)

[ 
https://issues.apache.org/jira/browse/YARN-1051?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13942122#comment-13942122
 ] 

Arun C Murthy edited comment on YARN-1051 at 3/20/14 6:50 PM:
--

More color on why I prefer priorities for reservations rather than 
adding/removing queues...

In vast majority of deployments, queues are an organizational/economic concept 
(e.g. per-department queues are very common) and are queues (hierarchy, names 
etc.) are quite stable and well recognized to point of being part of the 
institutional memory.

If we rely on adding/removing queues to provide reservations, I'm concerned it 
will cause some confusion among both admins and users. For e.g. a user/admin 
trying to debug his application will be quite challenged to figure 
demand/supply of resources when he has to go back in time to reconstruct a 
programmatically generated queue hierarchy, particularly after it's long gone.

Priorities, OTOH, is quite a familiar concept to admins (think unix 'nice'); 
and more importantly is a natural fit to the problem at hand i.e. temporally 
increase/decrease the priority of the application based on it's reservation at 
a point in time.

Furthermore, as I said previously, priorities are an often requested feature - 
especially by admins.


was (Author: acmurthy):
More color on why I prefer priorities for reservations rather than 
adding/removing queues...

In vast majority of deployments, queues are an organizational/economic concept 
(e.g. per-department queues are very common) and are queues (hierarchy, names 
etc.) are quite stable and well recognized and part of the institutional memory.

If we rely on adding/removing queues to provide reservations, I'm concerned it 
will cause some confusion among both admins and users. For e.g. a user/admin 
trying to debug his application will be quite challenged to figure 
demand/supply of resources when he has to go back in time to reconstruct a 
programmatically generated queue hierarchy, particularly after it's long gone.

Priorities, OTOH, is quite a familiar concept to admins (think unix 'nice'); 
and more importantly is a natural fit to the problem at hand i.e. temporally 
increase/decrease the priority of the application based on it's reservation at 
a point in time.

Furthermore, as I said previously, priorities are an often requested feature - 
especially by admins.

 YARN Admission Control/Planner: enhancing the resource allocation model with 
 time.
 --

 Key: YARN-1051
 URL: https://issues.apache.org/jira/browse/YARN-1051
 Project: Hadoop YARN
  Issue Type: Improvement
  Components: capacityscheduler, resourcemanager, scheduler
Reporter: Carlo Curino
Assignee: Carlo Curino
 Attachments: YARN-1051-design.pdf, curino_MSR-TR-2013-108.pdf, 
 techreport.pdf


 In this umbrella JIRA we propose to extend the YARN RM to handle time 
 explicitly, allowing users to reserve capacity over time. This is an 
 important step towards SLAs, long-running services, workflows, and helps for 
 gang scheduling.



--
This message was sent by Atlassian JIRA
(v6.2#6252)