From: Shobhit Raj Mathur (shobhit@cs.washington.edu)
Date: Tue Oct 05 2004 - 18:01:07 PDT
V.G. Cerf and R.E. Kahn, "A Protocol for Packet Network Interconnection"
------------------------------------------------------------------------
Main Result:
------------
The paper presents a protocol for communication between different packet switching networks. The design of the protocol
enables it to overcome the variations in the different individual networks and allows seamless intercommunication.
Strengths of the paper:
-----------------------
- The protocol suggested in the paper effectively tackles the problems involved in communication between different
packet switching networks by introducing the following ideas:
- Uniform addressing scheme which can be understood by each individual network.
- Allowing different individual network packet sizes.
- End to end restoration procedures to allow for recovery from transmission failures and errors.
- Controlling the rate of flow of data between the source and the receiver.
- The implementation of such a protocol could be varied, but the implementation chosen in this paper has the following
key ideas which make it simple and robust:
- Simple and reliable interface (Gateway) between individual networks.
- Assumption of a common protocol used between the processes in different networks.
- 'Fragmentation' of packets in a way that the destination is able to piece together the fragmented packets. This is
done by using headers for each segment which also contains useful flags(ES and EM).
- A Transmission Control Program (TCP) which handles transmission and acceptance of messages on behalf of processes
it serves. It is responsible for multiplexing and demultiplexing.
- Retransmission and duplicate detection are effectively handled using a simple window strategy.
- Communication between a process and TCP is done using buffers (TCB and RCB).
- A 'connection setup' is not required to synchronize the receiver with the sender.
- Simple accounting is possible based on packet sizes.
- This is the first paper to address the problem of intercommunication between different packet switching networks,
which makes the paper unique.
Key Limitations, Unproven Assumptions and Methodological Problems:
------------------------------------------------------------------
- Though the paper addresses flow control it fails to address the important issue of congestion control.
- The accounting procedure described is very basic. If a packet A moves from network X to network Y and is broken into
A1 and A2, Y charges X for the total size of A1+A2. This is faulty, as the extra cost of headers should not be paid
by X.
- The paper does not specify how the hosts are given distinct addresses using the uniform addressing scheme.
- More detailed information regarding how the Gateway does routing between different networks is required.
- In the section on "reassembly and sequencing", the notion of packets and segments are not introduced clearly. Though
it becomes clearer by the end of the section.
How could the work be improved ?:
---------------------------------
- Accounting should have been done on the basis of the text content in a segment rather than the size of the packet.
- Segments and Packets should have been introduced clearly.
- How the 16 bits of the TCP identifier are split between the address and port number should be specified.
- Typical timeout durations to be used by the sender before retransmitting would have been useful.
Relevance today and future work:
--------------------------------
The paper introduces the first TCP protocol which was used. It addresses the main issues of intercommunication between
different packet switching networks which are still valid 30 years later. The TCP protocol though modified in the later
years, still is the same fundamentally, which makes this a classic paper .
This archive was generated by hypermail 2.1.6 : Tue Oct 05 2004 - 18:01:08 PDT