[Planetlab-users] Why so high load averages?

[Vivek Pai]

Luis Garces-Erice wrote:
> Hello,
> sorry if I'm interferring, but i think this is very important for 
> anybody performing experiments on Planet-Lab, specially if s/he has 
> their publication in mind. I quote Steve:
> 
>  > It's a distributed network testbed, used by
>  > a large number of users, replete with all the accompanying issues.  If
>  > you're building a distributed application that assumes a 24/7 available
>  > system with the entire CPU available to one user then PlanetLab is
>  > probably not for you.
> 
> Well, perhaps Joe's is not a realistic assumption, but if I perform an 
> experiment, I want to control as much as possible the environmnet, to be 
> sure of what my results *really* mean. 

If you really don't want any variation in environment, I believe that
both Emulab and Modelnet can give you a more predictable test
environment. Not trying to drive anyone away from PlanetLab, but it's
more of a "right tool for the job" approach. If you want all of the
ugliness that really exists, PlanetLab is pretty good for that.

 > How can I be sure of a given
> performance measurement in Planet-Lab, if some machines might be hogged 
> ? How can I consider the results of a experiment if I'm not sure of 
> what's going on, and what might be affecting it ?

You can monitor the local environment in many ways and see what's
happening while your experiment is running. Not perfect, but it can
explain outliers.

-Vivek