[Planetlab-devel] [Fwd: Re: getting more info about ntp configuration]

[Y. Charlie Hu]

Marc suggested that I forward the issue here.

Basically, to assist us in picking suitable PlanetLab nodes
for a one-way delay study, we need to access the NTP configs on all the 
planetlab machines.  We are not asking them to be reconfigured or 
anything; all we need to know is which ntp server is configured on which 
machine.

Marc suggests perhaps the best way to do this is to give us the API
for retrieving /etc/ntp.conf? Or perhaps there is already an XMLRPC API 
for it?

Thanks,
Charlie





-------- Original Message --------
Subject: Re: getting more info about ntp configuration
Date: Thu, 31 May 2007 14:49:51 -0400
From: Marc E. Fiuczynski <mef at CS.Princeton.EDU>
To: Y. Charlie Hu <ychu at purdue.edu>

Hi Charlie,

I am at an event in Brazil with poor email/network connectivity.

This question would be better address on the devel mailing list.  Please
join devel at lists.planet-lab.org and send this message there, and someone
else from PL will be able to address this.

The contents of /etc/ntpd.conf is actually something that we obtain
programmatically via some PHP script running on our PLC server.  My
guess is that the best thing to do is give you the API to obtain this
information from the PHP script for every node. (It could be that we've
put this into our XMLRPC API, but I am not sure and cannot check from
down here).

For your one-way ping experiment you will likely also want to think
about the variable delay for receiving the packet on overloaded nodes.
I believe we associate timestamps with packets, but I am not sure
whether this is for incoming our outgoing packets.  This would be a good
thing to discuss on the PL users mailing list, as there are various
experts (e.g., Neil Spring @ U. Maryland --- aka Scriptroute) who have
done things like this or would know the difficulties of doing this on PL.

Marc


Y. Charlie Hu wrote:
> Hi Marc,
> 
> We are doing a project that studies the asymmetry
> in Internet one-way delays using the PlanetLab nodes.
> The tool we are using, oneway-ping, critically relies on
> synchronization between each pair of end hosts being measured,
> Unfortunately the ntp used for syncing these nodes
> itself suffers poor accuracy from delay asymmetry problem (between ntp 
> client and ntp server). So we are stuck in a "chicken-and-egg" problem.
> 
> One way out of the dilemma is to use planetlab nodes
> that are close to the ntp servers they use, which
> is often not the case. For example, with the help of our sysadmin,
> I found out the ntp servers used by planet{1,2}.cs.purdue.edu can be up 
> to 200ms away.
> 
> So we could really use your help with this --
> Is there a way for you to tell us the /etc/ntp.conf from each and every
> PlanetLab node somehow? I understand this file is not visible from
> an ordinary slice. The only way we will make use of this info
> is to identify nodes that fit the criteria above for use in our 
> experiments. We will not abuse it in anyway.
> 
> I greatly appreciate your help,
> Charlie
>