CS 535, Telecommunications

Quiz 1, 22 September 2004

  1. Isenberg identifies underspecification as one of the advantages of stupid networks over intelligent ones, and then identifies user control as a consequence of that advantage. What would a network look like if it had one without the other? That is, describe a) a network that has underspecification but not user control and b) a network that had user control but not underspecification. Your answer should be in terms of network properties resulting directly from the presence or absence of underspecification and user control.

    Isenberg relates "underspecification" to the range of behavior not explicitly required of the network. The broader the range of such behavior, the more underspecified the network is.

    User control" is a measure of the user's ability to act through the network. User control is inversely proportional to the degree to which the network interface constrains a user's actions; fewer constraints means more user control.

    An underspecified network without user control would be difficult to use. Because the network is underspecified, it can engage in a broad range of behaviors, most of which are probably counterproductive to the way in which the user wants the network to behave. If the user's actions are constrained by the network interface, then the user may find the network unusable because it's forcing the user to take actions that the network interface is preventing.

    A specified network (a not-underspecified network) with user control is a bit redundant. Although user control allows a broad range of actions on the other side of the interface, within the network those actions must conform to the network's specified behavior.

  2. Soft state is state that's kept within the network that is used to implement some functions for the endpoints. It's called "soft state" because it's not required to keep the network functioning; it just helps the network run better. Some people claim that adding soft-state to a network doesn't violate the end-to-end guideline. Discuss the pros and cons of that claim.

    The pros of soft state are given in the question: the network can exploit soft state to provide (perhaps) more effective service with (perhaps) greater network-operating efficiency.

    The cons are given by the motivations for the end-to-end guideline. The application can't relay on soft-state assist within the network, and so must still provide an end-to-end implementation of the required functions. Adding soft state to the network increases its design, implementation, and operating costs, with the hope that operating costs can be covered by the greater efficiencies shown by the applications using soft-state assist.

    Two other related observations could be either pro or con, depending on your point of view. To cover the costs of wide-spread implementation in the network, the soft-state assist would have to be applicable to broad range of applications. Failing that, applications that don't need the soft-state function, or can't use the soft-state function in the form provided by the network, will be paying for a function that they're not using. My view is towards the con side, but maybe that's just because I'm not smart enough to discover such universal soft-state functions.

This page last modified on 14 November 2004.