Transport protocols and bulk file transfer

[Lisong Xu <>]

> > 2) About the time precision of bandwidth estimation and the
> > delay-based protocols. I understand that bandwidth estimation may
> > benefit from a high precision. But I am not sure why
> > bandwidth-estimation require higher time precision than delay-based
> > protocols, since both of them are based on the delay changes which
> > are "on the same order of magnitude as the serialization delay at
> > the bottleneck".
>
>
> Suppose that we have a way to measure delay with an accuracy of 1ms
> (after filtering and whatever statistical techniques are to be used).
> It would seem like bandwidth-estimation techniques would not find that
> very useful.  At the same time, this should be usable to drive the
> target queue size to, say, 5ms, shouldn't it?
Thanks. I think what you said is correct for some bandwidth estimation 
techniques. But there are some protocols using the difference between 
the current RTT (or the current one-way delay) and the minimum RTT (or 
the corresponding min one-way delay). For example TCP-westwood, and 
HSTCP-LP. In this case, we may not need a high time precision.

As Constantinos is also on the list, and he is an expert on bandwidth 
estimation, maybe we can get some comments from him.

Thanks
Lisong