The following rules are applied when a lock conflict occurs with the MWDL method.
We summarize here the results of a simulation study with the restart waiting-no lock resampling options (justified in the previous section) and exclusive lock requests to compare the performance of eight methods: GW, NW, CWS, RPS and RPA, WDL, MWDL, and WW [Thomasian 1997].
For smaller values of P, RPA outperforms not only RPS but also the WDL and MWDL methods, which is attributable to the fact that RPA results in less wasted processing than the other methods, such that it can maximize the useful processing in the system.
1992] and the MWDL [Thomasian 1992] methods outperform others (including RPA) at higher processing capacities.
However, as P is increased, WDL and MWDL utilize more processors (up to P = 500) and attain a higher [P.sub.u] than RPA [Thomasian 1993).