Lottery Scheduling Review

From: Joanna Muench (joannam_at_spro.net)
Date: Tue Jan 27 2004 - 21:17:02 PST

  • Next message: Steve Arnold: "Review: Waldspurger & Weihl, Lottery Scheduling"

    Waldspurger and Weihl (1994) present an interesting solution to resource
    allocation via a lottery system. The authors successful demonstrate the
    current paucity of mechanisms to distribute resources flexibly but
    fairly. They primarily position their work as a solution to
    computational resources, but illustrate how to apply the mechanism to
    I/O bandwidth, memory and access to locks. The authors argue that their
    allocation algorithm is both fair and considerably simpler than
    previously proposed mechanisms. The later claim is easy to believe given
    how readable this paper is.

    The central idea to the lottery schedule is the allocation of tickets
    among clients, with each ticket within the ticket pool or list
    representing an equal chance of winning the resource. Tickets can be
    apportioned among the competing clients according to a prioritization
    scheme. The authors don't explicitly state that once used a ticket is
    returned to the ticket list, but that seems to be implicit in their
    argument. I believe this makes their mechanism equivalent to
    sampling-with-replacement, a well-established statistical method.

    I found the initial lottery mechanism quite interesting, but especially
    appreciated their discussions of topics such as ticket transfers and
    compensation tickets. The ticket transfer idea seemed particularly
    slick, as a way for a high priority task to transfer its priority to a
    service. The compensation ticket concept is a little clunky, but
    obviously necessary for processes that only run briefly.

    I would have liked to see more discussion about the problem of ticket
    inflation and protection mechanisms to prevent clients from generating
    extra tickets. The mechanism would need to be very flexible since
    inflation/deflation could be useful as an allocation tool, but could
    also unbalance the system if utilized improperly.

    Overall this paper did an excellent job of covering the concept of
    lottery scheduling from the abstract to several concrete examples. The
    authors provide an elegant and flexible mechanism to allocate resources
    preferentially with low-overhead.


  • Next message: Steve Arnold: "Review: Waldspurger & Weihl, Lottery Scheduling"

    This archive was generated by hypermail 2.1.6 : Tue Jan 27 2004 - 21:13:27 PST