[
https://issues.apache.org/jira/browse/MESOS-3765?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14971016#comment-14971016
]
Klaus Ma edited comment on MESOS-3765 at 10/23/15 2:09 PM:
-----------------------------------------------------------
Thanks for explaining current behaviour, it match my understanding :).
According to the description of this ticket, we consider assigning all
resources of slave to framework is unfair, right? So which case is fairness by
re-using the example?
Two framework {{f1}} and {{f2}}, one agent with only 1 CPU:
1. 0.5 to {{f1}} and 0.5 to {{f2}} (both weight are 1: {{total * weight / sum
of weight}})
2. 1 to {{f1}} and 0 to {{f2}}
3. or others?
IMO, #1 is fair to both framework; but if {{f1}}/{{f2}} acquired 1 CPU to
launch task, it need a way for allocator to adjust the fairness, so
{{requestResources()}} will help. And another option is to allow wast
resources: keep offering 0.5 to {{f1}} and 0.5 to {{f2}}.
was (Author: klaus1982):
Thanks for explaining current behaviour, it match my understanding :).
According to the description of this ticket, we consider assigning all
resources of slave to framework is unfair, right? So which case is fairness by
re-using the example?
Two framework {{f1}} and {{f2}}, one agent with only 1 CPU:
1. 0.5 to {{f1}} and 0.5 to {{f2}} (both weight are 1: {{total * weight/ sum
of weight}})
2. 1 to {{f1}} and 0 to {{f2}}
3. or others?
IMO, #1 is fair to both framework; but if {{f1}}/{{f2}} acquired 1 CPU to
launch task, it need a way for allocator to adjust the fairness, so
{{requestResources()}} will help.
> Make offer size adjustable (granularity)
> ----------------------------------------
>
> Key: MESOS-3765
> URL: https://issues.apache.org/jira/browse/MESOS-3765
> Project: Mesos
> Issue Type: Improvement
> Components: allocation
> Reporter: Alexander Rukletsov
>
> The built-in allocator performs "coarse-grained" allocation, meaning that it
> always allocates the entire remaining agent resources to a single framework.
> This may heavily impact allocation fairness in some cases, for example in
> presence of numerous greedy frameworks and a small number of powerful agents.
> A possible solution would be to allow operators explicitly specify
> granularity via allocator flags. While this can be tricky for non-standard
> resources, it's pretty straightforward for {{cpus}} and {{mem}}.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
