Bug 824 - TCP should implement FastRecovery by default
TCP should implement FastRecovery by default
Status: RESOLVED FIXED
Product: ns-3
Classification: Unclassified
Component: internet
ns-3-dev
All All
: P5 normal
Assigned To: Josh Pelkey
:
Depends on:
Blocks:
  Show dependency treegraph
 
Reported: 2010-02-25 02:25 EST by Tom Henderson
Modified: 2010-12-23 18:22 EST (History)
2 users (show)

See Also:


Attachments

Note You need to log in before you can comment on or make changes to this bug.
Description Tom Henderson 2010-02-25 02:25:06 EST
Please see this ns-3-users message for an outline from Antti:

http://groups.google.com/group/ns-3-users/browse_thread/thread/7e0acbf1b540dcbc
Comment 1 Josh Pelkey 2010-02-25 12:45:48 EST
I believe TCP Tahoe (the flavor we are supposed to have in ns-3) does not support fast recovery.  I thought that Reno was essentially TCP Tahoe with fast recovery.  Maybe I am reading the discussion wrong.  My information on the basic TCP flavors comes from comes from the Kevin Fall / Sally Floyd paper: Simulation-based Comparison of Tahoe, Reno, and SACK TCP.
Comment 2 Tom Henderson 2010-02-25 14:58:27 EST
(In reply to comment #1)
> I believe TCP Tahoe (the flavor we are supposed to have in ns-3) does not
> support fast recovery.  I thought that Reno was essentially TCP Tahoe with fast
> recovery.  Maybe I am reading the discussion wrong.  My information on the
> basic TCP flavors comes from comes from the Kevin Fall / Sally Floyd paper:
> Simulation-based Comparison of Tahoe, Reno, and SACK TCP.

Yes, I think it is probably correct to say that Tahoe includes Fast Retransmit.  For a while, there were two variants of Tahoe in the wild; those with Fast Retransmit and those without.  See the TBIT paper "On Inferring Tcp Behavior" by Padhye and Floyd.  

But I think it is still OK to call what we have now "Tahoe".
Comment 3 Antti Mäkelä 2010-03-02 04:37:09 EST
(In reply to comment #2)
> Yes, I think it is probably correct to say that Tahoe includes Fast Retransmit.
>  For a while, there were two variants of Tahoe in the wild; those with Fast
> Retransmit and those without.  See the TBIT paper "On Inferring Tcp Behavior"
> by Padhye and Floyd.  
> 
> But I think it is still OK to call what we have now "Tahoe".

  Mostly I'm interested in this functionality since if I submit a paper on how TCP works in my particular network configuration the first reviewer with a clue will ask "Why are you running with such an ancient TCP variant, what about newer mechanisms". 

  (Network simulation cradle doesn't allow such in-depth tracing, unfortunately - goodput, sure, but not durations of pauses, e.g. when downloading a webpage, user will rather press cancel than wait out if the transmission stops for more than couple of seconds..)
Comment 4 Tom Henderson 2010-12-23 18:22:06 EST
changeset: a814f37d15bf