Implementation of gang-scheduling on workstation cluster

Hori, Atsushi; Tezuka, Hiroshi; Ishikawa, Yutaka; Soda, Noriyuki; Konaka, Hiroki; Maeda, Munenori

doi:10.1007/bfb0022291

Cited by 24 publications

(18 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In [10], SCore-D is introduced as a gang scheduler. In addition, we have added the system-call mechanism and have implemented I/O mechanisms and other operating system functions.…”

Section: System-call Mechanismmentioning

confidence: 99%

“…The overhead of PC cluster is much less than that of workstation cluster. We guess this dierence is dominated by a scheduling policy of the local Unix operating system [10].…”

Section: Gang Scheduling Overheadmentioning

confidence: 99%

“…We have developed a user-level parallel operating system, called SCore-D [10,9], for workstation clusters. The key idea enabling gang scheduling in SCore-D is network preemption.…”

Section: Introductionmentioning

confidence: 99%

“…Therefore development a parallel operating system at user-level is reasonable for parallel operating system research. SCore-D is a user-level parallel operating system on top of UNIX [10,9]. The runtime system of MPC++ has been developed for SCore-D and user programs.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Global state detection using network preemption

Hori¹,

Tezuka²,

Ishikawa³

1997

Job Scheduling Strategies for Parallel Processing

Self Cite

View full text Add to dashboard Cite

Gang scheduling provides shorter response time and enables interactive parallel programming. To utilize processor resources on interactive parallel programs, global state of distributed parallel processes should be detected. This problem is well-known as \distributed termination problem." In this paper, we propose a practical method to detect a global state of distributed processes. There are two key methods to detect a global state described in this paper. One is by network preemption and the other is by combining global state detection with gang scheduling. The overhead for the detection of a global state is negligible. To implement our scheme, we extend our gang scheduler, SCore-D, to an operating system by implementing a system-call mechanism. We conrmed that our proposed global state detection scheme is very useful on SCore-D.

show abstract

“…In [10], SCore-D is introduced as a gang scheduler. In addition, we have added the system-call mechanism and have implemented I/O mechanisms and other operating system functions.…”

Section: System-call Mechanismmentioning

confidence: 99%

“…The overhead of PC cluster is much less than that of workstation cluster. We guess this dierence is dominated by a scheduling policy of the local Unix operating system [10].…”

Section: Gang Scheduling Overheadmentioning

confidence: 99%

“…We have developed a user-level parallel operating system, called SCore-D [10,9], for workstation clusters. The key idea enabling gang scheduling in SCore-D is network preemption.…”

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Global state detection using network preemption

Hori¹,

Tezuka²,

Ishikawa³

1997

Job Scheduling Strategies for Parallel Processing

Self Cite

View full text Add to dashboard Cite

show abstract

“…These processes usually perform intra-node communication via shared memory areas, obtained using System V IPC mechanisms. In the context of fine-grained parallel applications, it is important that related processes are scheduled to execute simultaneously ("gang scheduling", [7,11,12]), so as to reduce the context switching overhead, minimize synchronization latencies and ensure that the parallel job makes sufficient progress. This is taken into account by our scheduling policy, which makes sure that the required number of processors is available for all related processes of an application to execute concurrently, if this application is to be scheduled on the next time quantum.…”

Section: Memory Bandwidth Aware Scheduling Policiesmentioning

confidence: 99%

Memory Bandwidth Aware Scheduling for SMP Cluster Nodes

Koukis

Koziris

13th Euromicro Conference on Parallel, Distributed and Network-Based Processing

View full text Add to dashboard Cite

show abstract

Adaptive parallelism in compiler-parallelized code

Hall¹,

Martonosi

1998

Concurrency: Pract. Exper.

View full text Add to dashboard Cite

As moderate‐scale multiprocessors become widely used, we foresee an increased demand for effective compiler parallelization and efficient management of parallelism. While parallelizing compilers are achieving success at identifying parallelism, they are less adept at predetermining the degree of parallelism in different program phases. Thus, a compiler‐parallelized application may execute on more processors than it can effectively use – a waste of computational resources that becomes more acute as the number of processors increases, particularly for systems used as multiprogrammed compute servers. This paper examines the dynamic parallelism behavior of multiprogrammed workloads using programs from the Specfp95 and Nas benchmark suites, automatically parallelized by the Stanford SUIF compiler. Our results demonstrate that even the programs with good overall speedups display wide variability in the number of processors each phase (or loop) can exploit. We propose and evaluate a run‐time system mechanism that dynamically adjusts the number of processors used by a compiler‐parallelized application, responding to observed performance during the program's execution. Programs can thus adapt processor usage as they run, responding both to poor parallelism within certain parts of their code, and also to heavy multiprogramming loads during the execution. This mechanism improves workload performance up to 33% over one‐at‐a‐time runs of the workload programs. © 1998 John Wiley & Sons, Ltd.

show abstract

Implementation of gang-scheduling on workstation cluster

Cited by 24 publications

References 11 publications

Global state detection using network preemption

Global state detection using network preemption

Memory Bandwidth Aware Scheduling for SMP Cluster Nodes

Adaptive parallelism in compiler-parallelized code

Contact Info

Product

Resources

About