1 2 3 4 Previous Next 58 Replies Latest reply: Mar 29, 2012 4:18 AM by gimbal2 RSS

    Datagram degradation

    898586
      Would anyone wish to hazard a guess or better still give a reason as to why DatagramSocket comms are noticeably degraded - (more dropped packets) - when the number of peers simply increases from 2 to 3?
        • 1. Re: Datagram degradation
          EJP
          More collisions on the network.
          Applications slower to read, due to contention, so packets dropped when socket receive buffers are full.
          • 2. Re: Datagram degradation
            898586
            Thanks.
            But . . what . . even when each client has a separate socket . . .? Please elaborate on the type of collisions.
            • 3. Re: Datagram degradation
              EJP
              Collisions on the network. There are no sockets on the network, they are only a figment of the endpoiints' imaginations.
              • 4. Re: Datagram degradation
                898586
                Where are these collisions taking place specifically? Presumably at an interface point, rather than reversing up the wire into the following packet?
                • 5. Re: Datagram degradation
                  796440
                  895583 wrote:
                  Where are these collisions taking place specifically? Presumably at an interface point, rather than reversing up the wire into the following packet?
                  "Collision" meaning multiple hosts are transmitting simultaneously, leading to multiple packets on the wire simultaneously.

                  http://en.wikipedia.org/wiki/CSMA/CD
                  http://en.wikipedia.org/wiki/Carrier_sense_multiple_access
                  • 6. Re: Datagram degradation
                    898586
                    Is this a condition that only applies to UDP packets?
                    • 7. Re: Datagram degradation
                      EJP
                      No. I shoud also have mentioned packets being dropped by routers due to overload.
                      • 8. Re: Datagram degradation
                        898586
                        TCP effectively drops no packets as far as the client is concerned - would that be fair to say? Interesting that you made no differentiation between client and server in terms of dropped packets, although I didn't mention where the problem was.
                        • 9. Re: Datagram degradation
                          EJP
                          TCP effectively drops no packets as far as the client is concerned - would that be fair to say?
                          It would be fairer to say that TCP detects missing bytes from the stream and gets them resent.
                          Interesting that you made no differentiation between client and server in terms of dropped packets
                          There is no differentiation to be made.
                          although I didn't mention where the problem was.
                          You don't know where the problem is.
                          • 10. Re: Datagram degradation
                            898586
                            You don't know where the problem is.
                            That's true; however it would seem to me less likely that there would be packet collisions going into the client over a unidirectional bytestream - (aka DatagramSocket) - than those that might occur serverside, where the circuitry is carrying heterogeneous packets. Would any packet be dropped - droppable even - once it left the last staging router?
                            Could you put me right on this . . ?
                            • 11. Re: Datagram degradation
                              EJP
                              All the packets go over the same network, and that's where the collisions occur. I've already said that. The only reason a client might be more exempt fom packet loss is that it probably only has one socket and can probably read it faster so its recv buffer is less likely to fill. It's not something I would bet the house on though.
                              • 12. Re: Datagram degradation
                                898586
                                If you can, please say what you mean by "same network"? Thanks.
                                • 13. Re: Datagram degradation
                                  EJP
                                  Maybe you could confide in us what makes you think it isn't the same network?

                                  There is a network between the client and the server. The server sends to the client over that network. The client sends to the server over that network. Other things happen on that network. All those things can result in packet loss.
                                  • 14. Re: Datagram degradation
                                    898586
                                    Should I therefore rather think in terms of the network being more of a particle-wave duality medium in nature, where packets are akin to particles, but which nevertheless have wavelike characteristics as analogue signals on the wire, with the consequent result that interference patterns occur, cancelling packets, and giving rise to the packet loss/ dropping? Or is this just my pure imagination?
                                    1 2 3 4 Previous Next