[Beowulf] Adding Latency to a Cluster Environment

[Bogdan Costescu]

On Fri, 13 Feb 2004 joshh at cs.earlham.edu wrote:

> Here is an irregular question. I am profiling a software package that runs
> over LAM-MPI on 16 node clusters [Details Below]. I would like to measure
> the effect of increased latency on the run time of the program.

It appears that in your setup MPI uses TCP/IP as underlying protocol.  
Latency is a fuzzy parameter in TCP/IP. Adding to something fuzzy gives a
fuzzy result.  So there :-)

Suggestions:
- modify the routines that make MPI calls to call instead some wrapper 
routines that do some thumb twiddling before making the MPI call; this 
requires modification of the program source
- modify the MPI routines (well, if you use an open-source MPI 
implementation) to insert some delay, then relink your binary if static
- modify the kernel source to insert some delays in the TCP path - pretty 
hard as TCP is very complex
- modify the network driver to insert some delays in the Tx or Rx packet
path; not very difficult, but might be leveled by the delays of TCP.

The kernel modifications have the disadvantage that they also require some 
way to change the delay value, so adding a /proc entry, an ioctl, etc. 
unless you want to recompile the kernel and reboot after each delay 
change.

> For more details on these clusters follow the link below:
> http://cluster.earlham.edu/html/

Please tell to whoever coded that page that Opera doesn't display it 
properly. And I use Opera all the time ;-)
The page also doesn't specify an important detail: the network cards/chips 
used in the clusters.

-- 
Bogdan Costescu

IWR - Interdisziplinaeres Zentrum fuer Wissenschaftliches Rechnen
Universitaet Heidelberg, INF 368, D-69120 Heidelberg, GERMANY
Telephone: +49 6221 54 8869, Telefax: +49 6221 54 8868
E-mail: Bogdan.Costescu at IWR.Uni-Heidelberg.De


_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf