In this paper we consider the problem of scheduling precedence task graphs in
parallel processing where there can be disturbances in computation and
communication times. Such a phenomenon often occurs in practice, due to our
inability to exactly
predict the time because of system intrusion like cache miss
and packet transmission time in mediums like ethernet etc.
We propose a method based on the addition of some extra edges to protect the
initial scheduling from performing badly
due to such changes and provide an upper bound on the
performance guarantee for the scheduling algorithms. Moreover, this
construction guarantees a result at least as good
as the result obtained for the initial static scheduling.
We also show that this construction
is a minimal set in context of partially on-line scheduling.