5cf8326f   
   From: grahn+nntp@snipabacken.se   
      
   On Thu, 24 Sep 2009 07:22:27 -0700 (PDT), David Schwartz   
    wrote:   
   > On Sep 23, 8:03 am, Ouyang  wrote:   
   >   
   >> Yes, this specification only means HTTP won't do reliable check, so   
   >> the underlying layer better take care of this issue.   
   >> If underlying protocol is not reliable (like UDP), when the network is   
   >> of congestion, the HTTP layer will receive incomplete reqeust/   
   >> response, and HTTP layer   
   >> can not order the other endpoint to retransmit it (retransmit   
   >> mechanism is provided by underlying layer, assumed by HTTP),   
   >> Hence the only thing left to do for HTTP layer is to drop this request   
   >> if incomplete.   
   >   
   > How would it know the request was incomplete? UDP packets don't have   
   > any sequence indicator either.   
   >   
   > You simply *cannot* layer HTTP on top of UDP. There *must* be some   
   > intermediary protocol.   
      
   Or just enumerate the features of TCP, put a checkmark on those which   
   HTTP needs and UDP doesn't provide. There will be lots of them, and   
   you'd have to implement it yourself in that intermediary protocol. I   
   suspect it will look a lot like TCP.   
      
   The only way you can gain anything is if you can find a major TCP   
   feature which HTTP doesn't need.  I'm not sure what that would be,   
   if you intend to support anything from tiny GETs to gigabytes of data   
   across an imprefect network ...   
      
   /Jorgen   
      
   --   
     // Jorgen Grahn    O  o   .   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)