Truong, H-L and Fahringer, T.: Performance Analysis for MPI Applications with SCALEA. In Proc. of the 9 th European PVM/MPI Conf., Linz, Austria (September 2002).  Home/Search   Document Details and Download   Summary   Related Articles

....of phase j and on the effect of these phase imbalances. 3.2 Classification of the causes of blocking Blocking can be caused by all phases of the execution profile. These include the computation, the overheads and the message delays. At the highest level, we identify 4 major classes [2] 6][10]: 1. Communication, also called interprocess interaction or information movement: the overhead due to the exchange of data between processes, the time that does not overlap with the computation, in the sense of lost processor cycles [4] 2. Message delays: the time for the messages to traverse ....

....visual tool XPvm [5] which automatically measures the computation, communication and blocking phases of message passing parallel programs. Our tool extends this analysis by letting the programmer instrument its code to differentiate the different logical parts of the parallel program, as SCALEA [10]. Starting from the recorded detailed execution profile, we implemented an algorithm that calculates the causes of the blocking phases. process 1 process 2 process 3 Fig. 3. Algorithm for determining the cause of blocking T2 The algorithm handles the blocking phases in chronological order. A ....

Truong, H-L and Fahringer, T.: Performance Analysis for MPI Applications with SCALEA. In Proc. of the 9 th European PVM/MPI Conf., Linz, Austria (September 2002).

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC