Review of "End-To-End Arguments in System Design"

From: Michelle Liu (liujing@u.washington.edu)
Date: Tue Oct 12 2004 - 20:43:31 PDT

  • Next message: Lillie Kittredge: "e2e"

    Review of "End-To-End Arguments in System Design"

    Jing Liu

     

        This paper presents a design principle which is called end-to-end argument for system design. Network designers need to design layered communication protocols. Whether certain functions should be implemented in the lower levels or higher levels is crucial to the design.

        The author declares that the function placement has to consider performance, efficiency and application requirement. The proposed end-to-end argument says that the application has to involve in providing the function. It is impossible that by low level system itself this function is provided. The author shows many application examples to prove this argument. First, if lower level provides the function and many other applications do not need this function, this causes inefficiency. Second, low level system is lack of information and the whole picture of the application, so the performance of the high level applications might not be ensured. Third, if some functions are implemented in the low level, some applications won't work. For instance, for the real-time voice communication, if the retransmission is functioned by the low level, the delay is dramatic and this will make real-time voice communication impossible.

        Another good point of this paper is that the examples showed by the author almost cover the whole parts of network system design, such as error recovery, security, delivery acknowledgment, etc. This shows that the end-to-end argument is applicable in the whole network system design.

        One of the weaknesses of the paper is that after reading the paper, people would know that the end system needs to participate in implementing certain functions. However, what the low level system should implement is still not clear. There is no detailed guide for the placement of functions.

        This paper shows a principle for today's network design. When we implement certain functions, we have to consider performance and efficiency and then decide the placement of the function.


  • Next message: Lillie Kittredge: "e2e"

    This archive was generated by hypermail 2.1.6 : Tue Oct 12 2004 - 20:43:45 PDT