RE: [P2PSIP] P2PSIP diagnostics: PING discussion

[Song Haibin <melodysong@xxxxxxxxxx>]

Hi Das,

 

The slides you provided are interesting. As for the RTT, we may not need the timestamp. However, from the overlay management perspective, one may want to measure the one way delay from an initiator node to the destination node.  Then do you have any suggestions for that?

 

Best Regards,
Song Haibin
Email: melodysong@xxxxxxxxxx
Skype: alexsonghw

---

From: p2psip-bounces@xxxxxxxx [mailto:p2psip-bounces@xxxxxxxx] On Behalf Of Das, Saumitra
Sent: Wednesday, November 12, 2008 5:31 PM
To: p2psip@xxxxxxxx
Cc: ietf@xxxxxxxx
Subject: Re: [P2PSIP] P2PSIP diagnostics: PING discussion

 

Hi Song,

  Even in the planetlab testbed, the following paper at PAM 2008 ( http://pam2008.cs.wpi.edu/slides/pathak.ppt ) shows that more than 40% of nodes have more than 20ms NTP error. In a general overlay we are likely to find a larger fraction of nodes with error in their NTP time (since the planetlab testbed is managed by the people who own the machines while general overlay nodes are unlikely to be that well maintained). Unless NTP refresh is made mandatory within a p2psip implementation, relying on ntp would be unlikely to be helpful to diagnostics

Thanks

Saumitra

www.saumitra.info

===============

Dear all,

In P2PSIP base protocol, Ping is used to test connectivity along a path.

However, connectivity quality can not be measured well without some useful

information, such as the timestamp and HopCounter. In p2psip diagnostics

draft version 03 http://tools.ietf.org/id/draft-zheng-p2psip-diagnose-03.txt

we extend the Ping message payloads with a few kinds of useful information

for connectivity quality check purposes. 

The initiator node receiving the Ping response should compute the overlay

hops to the destination peer for the statistics of connectivity quality from

the perspective of overlay hops.

About the Timestamp, we are not sure if the NTP time is a good candidate for

it, or we have other ways to describe it, or we don't need the timestamp at

all. But if NTP time is used in the overlay, then it is a good choice,

because every peer along the message path will know when the message is

generated, and it is easy for the peer along the path to calculate the

message latency. However many overlays may not be synchronized with the NTP

time. Any comments?

Best Regards,

Song Haibin

Email: melodysong at huawei.com

Skype: alexsonghw

 

_______________________________________________

Ietf@xxxxxxxx
https://www.ietf.org/mailman/listinfo/ietf