Article in Press
This article is currently in the Just Accepted phase. The final published version may have formatting changes or additional corrections.
Abstract
Abstract—We consider uniformization methods for a Markov reward model and analyze the impact of the use of pairwise summation on the accuracy, CPU time and storage consumption of the methods. The impact on the accuracy of the methods is found significant when the average number of non-zeros per row of the infinitesimal generator of the underlying continuous-time Markov chain (CTMC) is over 20 or so and tends to increase with the average number of non-zeros per row. The overhead in CPU time ranges between approximately 45 % and 88 %. Finally, the storage overhead is smaller than about 200 %. Taking all those findings into account, the use of pairwise summation is an alternative to be considered when the average number of non- zeros per row of the infinitesimal generator of the underlying CTMC is over 20 or so.
