Kacheong Poon writes:
 > Roch - PAE wrote:
 > 
 > >  > It is worth trying.  But I suspect that the delay may
 > >  > need to be dynamic and adaptive.  For example, if the app is
 > >  > doing a simple request/response type of transaction, the
 > >  > above will introduce the minimum delay for every request and
 > >  > response.
 > >  > 
 > > 
 > > Apps that uses multiple small writes for a requests then
 > > wait for a response benefit greatly from setting
 > > TCP_NODELAY as this would bypass the block in both networked and
 > > fused case.
 > 
 > 
 > Yes.  But I was not suggesting this case.  I meant just a simple
 > request/response protocol.  The request and response can be a
 > single write.  Nagle algorithm won't matter in the networked
 > case so there is no need to set TCP_NODELAY.  But the above delay
 > on read will show up in loopback case.  I was trying to point
 > out a possible side effect on some apps if the delay is not
 > adaptive.

Ok. sorry.

I think we can be much like Nagle.

The delay   can be inserted   between  2 successive outgoing
writes. The first  write wakes  up the  peer  (and starts  a
timer).  A read, if issued, cancels the timer.  But a second
write shortly  after a first one will  not cause a wakeup of
the peer; data is buffered.  All like Nagle. Only difference
is that instead of  waiting for an  ACK, we  wait for a  fix
timer.

 > 
 > -- 
 > 
 >                                              K. Poon.
 >                                              [EMAIL PROTECTED]
 > 
 > _______________________________________________
 > networking-discuss mailing list
 > [email protected]

_______________________________________________
networking-discuss mailing list
[email protected]

Reply via email to