[GE users] Memory quotas

reuti reuti at staff.uni-marburg.de
Tue Feb 10 16:41:49 GMT 2009


Hi,

Am 10.02.2009 um 16:22 schrieb davidecittaro:

> Hi, last mail, I swear, just to be sure I haven't missed anything :-)
>
>> E.g.:
>>
>> - default in "qconf -mc": 5GB
>>
>
> $ qconf -sc | grep h_vmem
> h_vmem              h_vmem     MEMORY      <=    YES
> YES        4G       0

OK


>> - maximum limit in queue definition per slot: 40GB
>
> $ for i in `qconf -sql`; do qconf -sq $i | grep h_vmem; done
> h_vmem                40G
> h_vmem                40G

OK


>> - Limit per node across all jobs on this particular node (either in
>> exechost definition or as an RQS): 54GB
>>
>> - Limit in total per user across all of his jobs in an RQS: 100 GB
>>
>
> $ qconf -srqs MemoryQuota
> {
>     name         MemoryQuota
>     description  Memory quota for users. Nobody can use more than 100
> Gb RAM \
>     and no more than 50 Gb per host
>     enabled      TRUE
>     limit        hosts {*} to h_vmem=50G
>     limit        users {*} to h_vmem=100G
> }

You have to make two RQS out of it. The first matching and granting  
rule will bypass all following rules in this set.


> If I've done things right each user submit a job and 4 Gb RAM are
> given as max ram available for that job. Plus each host has at maximum
> 50 Gb available to run jobs and each user has at max 100 Gb RAM all
> over the cluster... Now I've lost why I've set h_vmem=40G in queue

This is a limit per job. So noone can block a complete node with just  
one job. If you don't need it, you could set this to the default  
"INFINITY".


> definition... Doesn't that mean that each queue has at max 40 Gb RAM
> available?

No, it's per slot. (It would be per queue-instance, if you would  
define it under "complex_values" in the queue definition. If you have  
two queues per node, it would still be per queue-instance, hence  
oversubscription might occur. Therefore we can do it in the exechost  
definition or a proper RQS, which sums up the consumption across all  
queues per node.)

-- Reuti


> Isn't this in conflict with limit set in RQS?
>
> d
>
>
>
> Davide Cittaro
> davide.cittaro at ifom-ieo-campus.it
>
> ------------------------------------------------------
> http://gridengine.sunsource.net/ds/viewMessage.do? 
> dsForumId=38&dsMessageId=103216
>
> To unsubscribe from this discussion, e-mail: [users- 
> unsubscribe at gridengine.sunsource.net].

------------------------------------------------------
http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=103229

To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].



More information about the gridengine-users mailing list