Under and over sharing equations relating throughput capacity, resource-sharing level and the source of thrashing in multi-thread systems
Date
2002-03-12
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
We propose a new, non-empirical, sharing-level version of the
sharing equation, which relates throughput capacity to the level of resource
sharing in a system. This sharing-level version uses a sharing-level
independent variable, as opposed to a sharing-algorithm time variable in an
earlier version of the equation. This sharing-level version is more
fundamental, given that the resource-sharing level in a system is a more
fundamental performance parameter that is more easily measured. There has
to be two distinct equations, however. One is for the case of under sharing,
the under-sharing equation, where the sharing level is such that the
throughput is less than the peak possible with an optimum sharing level. With
the under-sharing equation, throughput capacity climbs with increasing
sharing level. The other equation is for the case of over sharing, the
over-sharing equation, where the sharing level exceeds the optimum sharing
level. With the over-sharing equation, throughput capacity decreases with
increasing sharing level. The over-sharing equation also reveals the true
source of thrashing, when the system is operating under extreme over sharing
conditions. A single empirical sharing equation is also presented, which
works reasonably well for both over and under sharing, at the cost of a
somewhat less than true peak throughput capacity.
Description
Keywords
Computer Science