[Iccrg] ICCRG review of draft-sridharan-tcpm-ctcp-02

[[e-mail] (Lachlan Andrew)]

Greetings Saverio,

That is a good point, but the self-induced losses are similar to
standard TCP for typical buffer sizes.

Although we didn't include specific figures for retransmissions, Fig
9(a) our PFLDnet paper
<http://netlab.caltech.edu/lachlan/abstract/HybridPFLDnet08.pdf>
shows that it shares bandwidth equally with standard TCP, which
implies that it is not introducing more losses than another standard
flow would.

It's mechanism is different from purely loss-based proposals which are
simply more aggressive.  It is aggressive only when there is no
detectable delay.

If the buffer size is less than one or two packets per flow, it may be
slightly more aggressive.  However, if the flows all have low BDPs
then it won't be much more aggressive (i.e., similar retransmission
rate).  Conversely, if they have large BDPs then CTCP will get much
higher throughput than standard TCP, because standard TCP needs a
large buffer when the per-flow BDP is large.

Cheers,
Lachlan

2008/12/18 Saverio Mascolo <[e-mail]>:
> did someone measure goodput and retransmissions? TCPs using a more
> aggressive probing than VJ TCP usually seem to end up - in  good
> cases- with similar goodput but at the expense of larger packet
> retransmissions, which is not good!
> -saverio



-- 
Lachlan Andrew  Centre for Advanced Internet Architectures (CAIA)
Swinburne University of Technology, Melbourne, Australia
<http://caia.swin.edu.au/cv/landrew> <http://netlab.caltech.edu/lachlan>
Ph [phone]