[Planetlab-users] Traffic Restrictions in Tulip, Ping and Traceroute?

[Neil Spring]

Adrian,

On Sep 8, 2007, at 9:43 AM, Adrian Suter wrote:
> I am using ping and tulip on the planet-lab nodes to measure loss  
> rates. My question addresses mainly Planet-Lab system  
> administrators. In ping I will send 100 probes per measurement  
> (separated by default by 1 second).

Do you mean an experiment that lasts 100 seconds? (fine)  Or a ping  
that consists of 100 back-to-back probes every second? (likely not).

> Are there any restrictions in terms of number of ICMP (ping)  
> requests-replies per minute? I mean, how many different  
> destinations (using the above scheme) can I ping at the same time,  
> without bothering the Planet-Lab administrators?

It is an open question.  Typically, it is not the volume of  
measurements that becomes a problem (it can be, on *.uk machines  
watched by JANET administrators -- just avoid them), but rather the  
small headache of destination firewall watchers who believe that  
anything their firewall reports as strange is a malicious attack.  If  
you're keeping measurements within PlanetLab, destination firewalls  
won't be a problem.  If not, use a blacklist, and look around for the  
other best practices.

> Concerning tulip, I will send 50 to 100 probes per measurement,  
> i.e. I will use the --count option. As I am only interested in two  
> hops, I will also use the --start and --end options. Are there any  
> restrictions in terms of number of tulip-probes per minute? I mean,  
> how many different destinations (using the above scheme) can I  
> tulip :-) at the same time, without bothering the Planet-Lab  
> administrators?

I'm unclear what you want to do; I assume you mean the two hops near  
the destination.  I'd counsel toward avoiding the very last hop so  
that machines don't see so many repeated probes.   Even though tulip  
isn't aggressive in its probing, it isn't very quiet in that it needs  
many packets to estimate what it's going to estimate.

Scriptroute will cap transmission at 30KB/s, which is probably more  
than it will get out of most planetlab machines.   I don't think it  
will take too many concurrent tulip executions to exceed this rate  
limit.  The scriptroute daemon will slow transmission of probe trains  
to stay under the limit.

> And, what about traceroute? Are there any restrictions? I guess I  
> can't traceroute to 100 different destinations every 5 seconds...

As above, there are unwritten restrictions that no one knows how to  
follow.    If you find destinations that don't object to the  
tracerouting (by running locally, then from a few planetlab machines,  
then eventually in your experiment), I think 100 destinations every  
five seconds might be conservative.  Just be sure you know what  
you're getting out of traceroute before you do it (see the onelab  
traceroute paper or sidecar, if unfamiliar, both in IMC last year)  
and be sure that you can't just ask for someone else's topology data  
(see iplane at u washington).

I know I didn't answer your precise question on how to set the knobs  
for your own rate limits;  If you have the knobs, set them as  
conservatively as you can without compromising your experiment.

-neil