Ok, so if we/I have 48 nodes, each with 64 HT-cores and 128GB memory and I want 
YARN to FULLY utilize the whole system. My profile would/should be

cpu: 52 # 80% to core max
mem: 104448 # 80% of node mem

Or basically set it to yarn-site.xml yarn.nodemanager.resource.cpu-vcores and 
yarn.scheduler.maximum-allocation-mb


--
Jonathan (Bill) Sparks
Software Architecture
Cray Inc.

From: Darin Johnson <[email protected]<mailto:[email protected]>>
Date: Friday, June 12, 2015 9:06 AM
To: Bill Sparks <[email protected]<mailto:[email protected]>>
Cc: "[email protected]<mailto:[email protected]>" 
<[email protected]<mailto:[email protected]>>
Subject: Re: Myriad/YARN configuration

Bill,

I'm at a computer now so I'll take a little more time to explain.  The profiles 
are used by the REST API so that you can tell a the resource manager to add a 
new node manager with cpu and memory defined by the profile.  If you set a 
profile large as:
large:
  cpu: 16
  mem: 32768

then you could flexup using {profile: large, instances: 10} to get 10 
nodemanagers running with 32 GB memory and 16 cpu core capacity.  You could 
flexdown {instance: 10} to remove the capacity.  How to flexup/flexdown is 
documented in API.md.  You can also flexup/flexdown using the webui that runs 
in port 8192.

In the near future we plan to have fine grain scaling, then when you launch a 
resource manager you'll also launch instance of node manager with 0 cpu and 0 
mem capacity.  As you run yarn applications the node managers will add cpu and 
mem capacity to meet your jobs container needs.

Let me know if you have more questions.

Darin

On Thu, Jun 11, 2015 at 6:43 PM, Bill Sparks 
<[email protected]<mailto:[email protected]>> wrote:
I've changed them with no ill effect. Just wondering what the purpose is. My 
yarn configurations will allow much more that the profiles set by mesos/myriad. 
So is myriad scheduler going to bounce the request based on profiles or be 
passive and let yarn containers do the work. All this is early days for me.

--
Jonathan (Bill) Sparks
Software Architecture
Cray Inc.

From: Darin Johnson <[email protected]<mailto:[email protected]>>
Date: Thursday, June 11, 2015 5:36 PM
To: "[email protected]<mailto:[email protected]>" 
<[email protected]<mailto:[email protected]>>
Cc: Bill Sparks <[email protected]<mailto:[email protected]>>
Subject: Re: Myriad/YARN configuration

I've changed the profiles before with no ill effect, you should be fine.

On Thu, Jun 11, 2015 at 6:28 PM, Adam Bordelon 
<[email protected]<mailto:[email protected]>> wrote:
+dev@myriad, -user@mesos

On Thu, Jun 11, 2015 at 8:51 AM, Bill Sparks 
<[email protected]<mailto:[email protected]>> wrote:

>  I have a question about the profiles configured
> in myriad-config-default.yml, do they have to match parameters specified in
> yarn-site.xml.
>
>  E.g.
> myriad-config-default.yml:
>    large:
>     cpu: 4
>     mem: 4096
>
>  yarn-site.xml:
>  <property>
>     <name>yarn.nodemanager.resource.cpu-vcores</name>
>     <value>${nodemanager.resource.cpu-vcores}</value>
> </property>
>
>  <property>
>     <name>yarn.nodemanager.resource.memory-mb</name>
>     <value>${nodemanager.resource.memory-mb}</value>
> </property>
>
>  I ask because we have large memory/high core count nodes 48 cores/64GB
> memory.
>
>  --
>  Jonathan (Bill) Sparks
> Software Architecture
> Cray Inc.
>


Reply via email to