[GE users] clusters on a subnet

Mark Casella mark at exa.com
Wed Apr 12 17:47:29 BST 2006


    [ The following text is in the "ISO-8859-1" character set. ]
    [ Your display is set for the "ISO-8859-10" character set.  ]
    [ Some special characters may be displayed incorrectly. ]

Hi Craig,

Our compute nodes are on separate private subnets, both from each other 
and from the rest of the organizational network.  The head node of each 
cluster acts as a NATing router for the cluster.  There is a single 
/home filesystem for all clusters.

Thanks for your feedback,

Mark

Craig Tierney wrote:
> Mark Casella wrote:
> 
>> Hi everyone,
>>
>>     I've been using 5.3p6 for over a year now with great results. I 
>> seem to learn something new everyday though :)
>>
>>     In house we recently got 2 new Linux clusters and they were put on 
>> a subnet so that the qmaster can not directly rsh to each of the 64 
>> nodes. You must first log into a mgmt node for that cluster and then 
>> you can rsh to all the nodes. All of the compute nodes have 
>> non-routable ip addresses (because we were all out of ip addresses...).
>>
>>     I read about transfer queues, but this doesn't seem to be exactly 
>> what I need, I would like ideally to just have the qmaster communicate 
>> to the execd's through the mgmt node. Also I would prefer to just have 
>> one qmaster to centralize all the accounting and usage info.
>>
>>     Has anyone run across this before? Are we better off finally 
>> making the leap to 6.0 (maybe this is a non-issue for 6.0, I read that 
>> port 536 is not used for commd in 6.0? how does it communicate then?)
>>
>>     If the simplest option is to just obtain more ip addresses for all 
>> the compute nodes, that's fine, I just wanted to check here first.
>>
> 
> Are your compute nodes on private subnets or public subnets relative to
> the rest of the department or organization?  I think you are best off
> renumbering the nodes.   If you want one qmaster for all of the 
> clusters, transfer queues seem to be a bit overkill.
> 
> You could add additional ethernet interfaces to the qmaster and connect
> them directly to the new subnets.  However, I don't know if this is
> how you want to run the system.  Without more information about the
> network architecture as well as how users access the systems and is
> there a single /home filesystem for all clusters, it is hard to see
> another solution.
> 
> Craig
> 
> 
> 
>> Thanks for any tips or advice,
>>
>> Mark
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
>> For additional commands, e-mail: users-help at gridengine.sunsource.net
>>
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> For additional commands, e-mail: users-help at gridengine.sunsource.net
> 
> 

-- 
Mark Casella
Principal Applications Engineer
Exa Corporation
Phone:	(781) 676-8575
Fax:	(781) 676-8599
http://www.exa.com

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list