Fabrizio Petrini. Network Performance with Distributed Memory Scientic Applications. Submitted to the Journal of Parallel and Distributed Computing, September 1998. Home/Search Document Details and Download
Summary Related Articles Check |
This paper is cited in the following contexts:
A New Approach to Parallel Program Development and.. - Petrini, Bassetti.. (1999) Self-citation (Petrini) (Correct)
....The vast majority of scienti c applications display bursty communication patterns with alternating spikes of impulsive communication with periods of inactivity. In Figure 2, we show the network utilization by running four distinct scienti c applications over a parallel machine with 256 processors [10]. In all four cases, we can identify communication holes, i.e. periods of network inactivity, in the network. Therefore, there exists a signi cant amount of communication bandwidth which can be made available for other purposes. 0 10 20 30 40 50 60 70 80 90 0 10000 20000 30000 40000 ....
Fabrizio Petrini. Network Performance with Distributed Memory Scientic Applications. Submitted to the Journal of Parallel and Distributed Computing, September 1998.
Scheduling with Global Information in Distributed Systems - Petrini, Feng (2000) (2 citations) Self-citation (Petrini) (Correct)
....network. The vast majority of parallel applications display bursty communication patterns with alternating spikes of impulsive communication with periods of inactivity. In Figure 2, we show the network utilization by running four distinct applications over a parallel machine with 256 processors [26]. In all four cases, we can identify communication holes, i.e. periods of network inactivity, in the network. Therefore, there exists a signi cant amount of communication bandwidth which can be made available for other purposes. 7 3 Multitasking Parallel Jobs In order to improve the resource ....
Fabrizio Petrini. Network Performance with Distributed Memory Scientic Applications. Submitted to the Journal of Parallel and Distributed Computing, September 1998.
Improved Resource Utilization with Buered Coscheduling - Petrini, Feng Self-citation (Petrini) (Correct)
....in Section 3, and some preliminary results are presented in Section 4. Finally, we present our conclusions in Section 5. 2 Resource Utilization of Parallel Programs In Figure 1, we show the network utilization by running four distinct applications over a parallel machine with 256 processors [22]. In all four cases, we can identify communication holes, i.e. periods of network inactivity, in the network. The processors are connected with an indirect interconnection network using state of the art routers. Based on these gures, there is obviously an uneven and inecient use of system ....
....in a parallel machine. Another important characteristic shared by many parallel programs is their access pattern to the network. The vast majority of parallel applications display bursty communication patterns with alternating spikes of impulsive communication with periods of inactivity [22]. Thus, there exists a signi cant amount of communication bandwidth which can be made available for other purposes. 3 Bu ered Coscheduling To improve the resource utilization of parallel programs, we propose to multitask parallel jobs. That is, instead of overlapping computation with ....
Fabrizio Petrini. Network Performance with Distributed Memory Scientic Applications. Submitted to the Journal of Parallel and Distributed Computing, September 1998.
Efficient Scheduling of Parallel Jobs on Massively Parallel.. - Fabrizio Petrini And Self-citation (Petrini) (Correct)
....Utilization in a Transpose FFT. Another important characteristic shared by many parallel programs is their access pattern to the network. The vast majority of parallel applications display bursty communication patterns with alternating spikes of impulsive communication with periods of inactivity [15]. Thus, there exists a significant amount of communication bandwidth which can be made available for other purposes. 3 Buffered Coscheduling To improve the resource utilization of parallel programs, we propose to multitask parallel jobs. That is, instead of overlapping computation with ....
Fabrizio Petrini. Network Performance with Distributed Memory Scientific Applications. Submitted to the Journal of Parallel and Distributed Computing, September 1998.
Buffered Coscheduling: A New Methodology for Multitasking.. - Petrini, Feng (2000) (6 citations) Self-citation (Petrini) (Correct)
....in an FFT Transpose Algorithm. Another important characteristic shared by many parallel programs is their access pattern to the network. The vast majority of parallel applications display bursty communication patterns with alternating spikes of impulsive communication with periods of inactivity [13]. Thus, there exists a significant amount of unused network bandwidth which could be used for other purposes. 2.2 Communication Buffering Instead of incurring communication and scheduling overhead on a per message basis, we accumulate the communication messages generated by each processor and ....
F. Petrini. Network Performance with Distributed Memory Scientific Applications. Submitted to the Journal of Parallel and Distributed Computing, September 1998.
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