Effective Selection of Partition Sizes for Moldable Scheduling of Parallel Jobs

Srinivasan, Srividya; Subramani, V.; Kettimuthu, Rajkumar; Holenarsipur, Praveen; Sadayappan, P.

doi:10.1007/3-540-36265-7_17

Cited by 33 publications

(26 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our idea is similar to this, but the difference is that our system does not rely on a set of possible requests from job users but makes molding decisions on its own within a certain limitation. In [15], another moldable job scheduling scheme was proposed, whose performance was not very robust as it degraded quite significantly under varying conditions of load and job scalability. To overcome this problem, S. Srinivasan [16] proposed enhancements to the moldable scheduling strategy that provided more reliable performance under many circumstances.…”

Section: Job Schedulingmentioning

confidence: 99%

Job Mapping and Scheduling on Free-Space Optical Networks

Fujiwara

Koibuchi

2016

IEICE Trans. Inf. & Syst.

View full text Add to dashboard Cite

SUMMARYA number of parallel applications run on a highperformance computing (HPC) system simultaneously. Job mapping and scheduling become crucial to improve system utilization, because fragmentation prevents an incoming job from being assigned even if there are enough compute nodes unused. Wireless supercomputers and datacenters with free-space optical (FSO) terminals have been proposed to replace the conventional wired interconnection so that a diverse application workload can be better supported by changing their network topologies. In this study we firstly present an efficient job mapping by swapping the endpoints of FSO links in a wireless HPC system. Our evaluation shows that an FSOequipped wireless HPC system can achieve shorter average queuing length and queuing time for all the dispatched user jobs. Secondly, we consider the use of a more complicated and enhanced scheduling algorithm, which can further improve the system utilization over different host networks, as well as the average response time for all the dispatched user jobs. Finally, we present the performance advantages of the proposed wireless HPC system under more practical assumptions such as different cabinet capacities and diverse subtopology packings.

show abstract

Section: Job Schedulingmentioning

confidence: 99%

Job Mapping and Scheduling on Free-Space Optical Networks

Fujiwara

Koibuchi

2016

IEICE Trans. Inf. & Syst.

View full text Add to dashboard Cite

show abstract

“…Moldable or malleable applications [17] can be directed to trade resources for execution time, for example, an application may run on 4 cores for 1 hour or 2 cores for 2 hours. This additional flexibility in resource management has been shown to bring significant gains [25,27,44,47]. However, the amount of computations essentially remains the same.…”

Section: Application Vs Infrastructure Centricmentioning

confidence: 99%

Introducing service-level awareness in the cloud

Klein

Maggio

Årzén

et al. 2013

Proceedings of the 4th Annual Symposium on Cloud Computing

View full text Add to dashboard Cite

Abstr actResource allocation in clouds is mostly done assuming hard requirements, applications either receive the requested resources or fail. Given the dynamic nature of workloads, guaranteeing on-demand allocations requires large spare capacity. Hence, one cannot have a system that is both reliable and efficient.To solve this issue, we introduce Service Level (SL) awareness in clouds, assuming applications contain some optional code that can be dynamically deactivated as needed. First, we design a model for such applications and synthesize a controller to decide when to execute the optional code and when to skip it. Then, we propose a Resource Manager (RM) that allocates resources to multiple SL aware applications in a fair manner. We theoretically prove properties of the overall system using control and game theory.To show the practical applicability, we implemented SL aware versions of RUBiS and RUBBoS with less than 170 lines of code. Experiments show that SL awareness may enable a factor 8 improvement in withstanding flash-crowds or failures. SL awareness opens up more flexibility in cloud resource management, which is why we encourage further research by publishing all source code. Keywor dsResource allocation, Service Level ABSTRACTResource allocation in clouds is mostly done assuming hard requirements, applications either receive the requested resources or fail. Given the dynamic nature of workloads, guaranteeing on-demand allocations requires large spare capacity. Hence, one cannot have a system that is both reliable and efficient.To solve this issue, we introduce Service Level (SL) awareness in clouds, assuming applications contain some optional code that can be dynamically deactivated as needed. First, we design a model for such applications and synthesize a controller to decide when to execute the optional code and when to skip it. Then, we propose a Resource Manager (RM) that allocates resources to multiple SL aware applications in a fair manner. We theoretically prove properties of the overall system using control and game theory.To show the practical applicability, we implemented SL aware versions of RUBiS and RUBBoS with less than 170 lines of code. Experiments show that SL awareness may enable a factor 8 improvement in withstanding flash-crowds or failures. SL awareness opens up more flexibility in cloud resource management, which is why we encourage further research by publishing all source code.

show abstract

“…However, the cited work assumes that applications have constant area (i.e., doubling the number of hosts halves the execution time), which does not hold for many applications. Other works assume more realistic moldable application models [13] and study how to efficiently schedule them on super-computers [25], [24].…”

Section: A Theoretical Studies Of Moldabilitymentioning

confidence: 99%