Multiprogramming on multiprocessors

Crovella, Mark; Das, P.; Dubnicki, Cezary; LeBlanc, Thomas J.; Markatos, Evangelos P.

doi:10.1109/spdp.1991.218246

Cited by 38 publications

(19 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our study is leveraged by previous research in OS job scheduling on multiprogrammed SMMs [8,18,29 Table 9: Performance improvement of scheduler-conscious event waiting over simple spin-then-block on an SGI Origin 2000. 32,33].…”

Section: Conclusion and Related Workmentioning

confidence: 99%

“…It has been widely acknowledged in the OS community that space/time sharing is preferable, outperforming other alternatives such as pure co-scheduling or gang-scheduling for better job response times [8,18,29,32,33]. The modern operating systems such as Solaris 2.6 and IRIX 6.5 have adopted such a policy in parallel job scheduling (see a discussion in Section 6 on gang-scheduling used in the earlier version of IRIX).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Adaptive two-level thread management for fast MPI execution on shared memory machines

Shen

Tang

Yang

1999

Proceedings of the 1999 ACM/IEEE Conference on Supercomputing

View full text Add to dashboard Cite

This paper addresses performance portability of MPI code on multiprogrammed shared memory machines. Conventional MPI implementations map each MPI node to an OS process, which suffers severe performance degradation in multiprogrammed environments. Our previous work (TMPI) has developed compile/run-time techniques to support threaded MPI execution by mapping each MPI node to a kernel thread. However, kernel threads have context switch cost higher than user-level threads and this leads to longer spinning time requirement during MPI synchronization. This paper presents an adaptive two-level thread scheme for MPI to reduce context switch and synchronization cost. This scheme also exposes thread scheduling information at user-level, which allows us to design an adaptive event waiting strategy to minimize CPU spinning and exploit cache affinity. Our experiments show that the MPI system based on the proposed techniques has great performance advantages over the previous version of TMPI and the SGI MPI implementation in multiprogrammed environments. The improvement ratio can reach as much as 161% or even more depending on the degree of multiprogramming.

show abstract

Section: Conclusion and Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Adaptive two-level thread management for fast MPI execution on shared memory machines

Shen

Tang

Yang

1999

Proceedings of the 1999 ACM/IEEE Conference on Supercomputing

View full text Add to dashboard Cite

show abstract

“…It can be seen that, with this increased base priority, the latency program using spin-block message receipt with priority boosting on process wakeup achieves quite good performance, except that it seizes too large a share of the CPU in many cases, where DCS can be controlled more nely by manipulating its fairness parameters. 5 Because ,1 is the smallest possible priority increase available with nice, the latency program cannot be made more fair in the case without DCS.…”

Section: E Ects Of Boosting the Base Process Priority With Nicementioning

confidence: 99%

Demand-based coscheduling of parallel jobs on multiprogrammed multiprocessors

Sobalvarro

Weihl

1995

Job Scheduling Strategies for Parallel Processing

View full text Add to dashboard Cite

This thesis describes demand-based c oscheduling, a new approach t o s c heduling parallel computations on multiprogrammed multiprocessors. In demand-based coscheduling, rather than making the pessimistic assumption that all the processes constituting a parallel job must be simultaneously scheduled in order to achieve good performance, information about which processes are communicating is used in order to coschedule only these; the resulting scheme is well-suited to implementation on a w orkstation cluster because it is naturally decentralized. I present an analytical model and simulations of demand-based coscheduling, an implementation on a cluster of workstations connected by a high-speed network, and a set of experimental results. An analysis of the results shows that demand-based coscheduling successfully coschedules parallel processes in a timeshared workstation cluster, signi cantly reducing the response times of parallel computations. AbstractThis thesis describes demand-based c oscheduling, a new approach t o s c heduling parallel computations on multiprogrammed multiprocessors. In demand-based coscheduling, rather than making the pessimistic assumption that all the processes constituting a parallel job must be simultaneously scheduled in order to achieve good performance, information about which processes are communicating is used in order to coschedule only these; the resulting scheme is well-suited to implementation on a w orkstation cluster because it is naturally decentralized. I present an analytical model and simulations of demand-based coscheduling, an implementation on a cluster of workstations connected by a high-speed network, and a set of experimental results. An analysis of the results shows that demand-based coscheduling successfully coschedules parallel processes in a timeshared workstation cluster, signi cantly reducing the response times of parallel computations. Acknowledgments 1Sometimes when looking through a thesis I nd that it contains a very long acknowledgments section, which more often than not proves upon examination to be a very personal document. When the writer is a friend, I enjoy reading the acknowledgments, especially if my name is mentioned. When the writer is someone I know only professionally, m y reaction is typically embarassment tinged with distaste. I am reminded of a Peanuts cartoon I read when I wa s a c hild, in which Snoopy s a w a dog riding in a car with its head out the window and its tongue apping in the breeze. Snoopy thought that if he were in a car he would never behave in so undigni ed a manner; rather, he would sit up straight and wear a seat belt. It will be clear, then, that in order to maintain a digni ed posture and spare the reader embarassment, I should like to write a brief and formal acknowledgments section, but when I look back on the work that has occupied me for most of three years, I realize that that I am so indebted to colleagues, family, and friends, that to do so would be out of the question. Thus to the reader who does not kno...

show abstract

“…This has invited researchers to develop different techniques in an attempt to adapt the traditional uniprocessor time-shared scheduler to the new situation of mixing local and parallel workloads [1,10]. Basically, there are two methods of making use of these CPU idle cycles, task migration [6,7] and time-slicing scheduling [8,9]. In a NOW, in accordance with the research carried out by Arpaci [10], task migration overheads and the unpredictable behavior of local users may lower the effectiveness of this method.…”

Section: Introductionmentioning

confidence: 99%