[Swift-devel] Need pointers to BG/P info for Swift execution

[ZHAO ZHANG]

Hi, Mike

I am cc'ing Jon here. I am not sure how swift is currently configured on 
BG/P.
Here are some instructions to run stuff on there.

On 1/26/2012 3:33 PM, Michael Wilde wrote:
> Justin, Zhao, All,
>
> Its been a long time since I have personally run Swift scripts on the BG/Ps.
>
> I am trying to help Jon get started on the BG/P, but I have forgotten some of the basics. Can you point out the info a Swift BG/P user needs regarding:
>
> - how to make sure that workers and the swift app() programs they launch have the right Linux environment (ie, full bash, full env, not limited "busybox" tools)
Please find out in the .sh file in the attachment. To run it, execute 
"cqsub -p MTCScienceApps -q prod-devel -k zepto-vn-eval 
$PATH/cnip-start.sh" on either challenger or surveyor.
>
> - how to ssh/telnet to worker nodes
On surveyor and intrepid,
Run "cqstat | grep running" or "cqstat -f | grep running", then you get
467064  toussain  02:00:00  512    running      ANL-R03-M0-512

Running " /soft/apps/ZeptoOS/bin/listip ANL-R11-M0-512" returns the list 
of the IO nodes
172.16.5.9
172.16.5.10
172.16.5.11
172.16.5.12
172.16.5.13
172.16.5.14
172.16.5.15
172.16.5.16
Then you can ssh to those IO nodes.


On Challenger, it is a bit different,
Given a job status as following
467717  felker    00:30:00  64     running    CHR-R00-M1-N08-64
Running "nslookup R00-M1-N08-J00" returns the IP of the IO node. 
172.16.9.49 in this case.

On each IO node, it has 64 compute nodes with it. The ip address is from 
192.168.1.1 to 192.168.1.64.
 From the compute nodes' point of view, the IO node's address is 
192.168.1.254 .
We have to telnet to those compute nodes.

You may find the first couple of lines of cnip-start.sh is setting the 
IP on the compute nodes.
That is the IP on torus network, it is a global network across all 
compute nodes within a single allocation.

best
zhao

>
> - anything else different from a normal cluster?
>
> - tips for working with Cobalt
>
> Thanks,
>
> - Mike
>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: cnip-start.sh
URL: <http://lists.mcs.anl.gov/pipermail/swift-devel/attachments/20120127/1e2e42ec/attachment.ksh>