Scheduling parallel jobs on multicore clusters using CPU oversubscription

Utrera, Gladys; Corbalan, Julita; Labarta, Jesús

doi:10.1007/s11227-014-1142-9

Cited by 11 publications

(8 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Instead, P2PGrid relies on dynamic mechanisms for matching jobs to free resources. The utility of over-subscription in centralized scheduling has been noted in other work [26].…”

Section: Parallel Job Schedulingmentioning

confidence: 75%

Parallel Computing with P2P Desktop Grids

Jackson

2014

2014 14th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing

View full text Add to dashboard Cite

Tightly-coupled parallel computing is an important tool for problem solving.Structured peer-to-peer network overlays are failure-tolerant and have a low administrative burden. This work seeks to unite the two.First, I present a completely decentralized algorithm for parallel job scheduling and load balancing in distributed peer-to-peer environments. This algorithm is useful for meta-scheduling across known clusters and scheduling on desktop grids.To accomplish this, I build on previous work to route jobs to appropriate resources then use the new algorithm to start parallel jobs and balance load across the grid. I also discuss what constitutes useful clusterings for this algorithm as well as inherent scaling limitations. Ultimately, I show that my algorithm performs comparably to one using centralized load balancing with global up-to-date information. The principal contribution of this work is that the parallel job scheduling is completely decentralized, which is not featured in previous work, and enables reliable ad hoc sharing of distributed resources to run parallel computations.Second, I show how clusters of computers can be found dynamically by using an existing latency prediction technique coupled with a new refinement algorithm.Several latency prediction techniques are compared experimentally. One, based on a tree metric space embedding, is found to be superior to the others. Nevertheless, I show that it is not quite accurate enough. To solve this problem, I present a refinement algorithm for producing quality clusters while still maintaining bounds for the amount of information any given node must store about other nodes. Ishow that clusters derived this way have scheduler performance comparable to those chosen statically with global knowledge.Lastly, I discuss previously undiscovered under-specifications in the Content Addressable Network (CAN) structured peer to peer system. In high-churn situations, the CAN allows stale information and changes to the overlay structure to create routing problems. I show solutions to these two problems, as well as discuss other issues that may also disrupt a CAN.

show abstract

“…Instead, P2PGrid relies on dynamic mechanisms for matching jobs to free resources. The utility of over-subscription in centralized scheduling has been noted in other work [26].…”

Section: Parallel Job Schedulingmentioning

confidence: 75%

Parallel Computing with P2P Desktop Grids

Jackson

2014

2014 14th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing

View full text Add to dashboard Cite

show abstract

“…It has also been studied how bad placement of processes for checkpoint/restart may hurt performance [38]. Another work describes how, even if possibly harmful inside one application, oversubscribing can be used to efficiently execute multiple applications sharing one node [37]. To circumvent this drawback when applying MPI oversubscribing in a unique application, some work focused on enabling multiple MPI process in one OS process [26], verifying the positive impact of such implementation.…”

Section: Explicit Methodsmentioning

confidence: 99%

Checkpoint/restart approaches for a thread-based MPI runtime

Adam¹,

Kermarquer

Besnard³

et al. 2019

Parallel Computing

View full text Add to dashboard Cite

Fault-tolerance has always been an important topic when it comes to running massively parallel programs at scale. Statistically, hardware and software failures are expected to occur more often on systems gathering millions of computing units. Moreover, the larger jobs are, the more computing hours would be wasted by a crash. In this paper, we describe the work done in our MPI runtime to enable both transparent and application-level checkpointing mechanisms. Unlike the MPI 4.0 User-Level Failure Mitigation (ULFM) interface, our work targets solely Checkpoint/Restart and ignores other features such as resiliency. We show how existing checkpointing methods can be practically applied to a thread-based MPI implementation given sufficient runtime collaboration. The two main contributions are the preservation of high-speed network performance during transparent C/R and the over-subscription of checkpoint data replication thanks to a dedicated user-level scheduler support. These techniques are measured on MPI benchmarks such as IMB, Lulesh and Heatdis, and associated overhead and trade-offs are discussed.

show abstract

“…Oversubscription was used there to enable computation to overlap with communication effectively hiding network latency [24]. In a previous work, Utrera et al [25], proposed a mechanism, the Load Balancing Detector (LDB), to classify applications dynamically, without any previous knowledge of it, depending on their balance degree and apply the appropriate process queue type to each job. LDB demonstrated to work especially well for the imbalanced jobs.…”

Section: Related Workmentioning

confidence: 99%

Task Packing: Efficient task scheduling in unbalanced parallel programs to maximize CPU utilization

Utrera

Farreras

Fornes

2019

Journal of Parallel and Distributed Computing

Self Cite

View full text Add to dashboard Cite

Load imbalance in parallel systems can be generated by external factors to the currently running applications like operating system noise or the underlying hardware like a heterogeneous cluster. HPC applications working on irregular data structures can also have difficulties to balance their computations across the parallel tasks. In this article we extend, improve and evaluate more deeply the Task Packing mechanism proposed in a previous work. The main idea of the mechanism is to concentrate the idle cycles of unbalanced applications in such a way that one or more CPUs are freed from execution. To achieve this, CPUs are stressed with just useful work of the parallel application tasks, provided performance is not degraded. The packing is solved by an algorithm based on the Knapsack problem, in a minimum number of CPUs and using oversubscription. We design and implement a more efficient version of such mechanism. To that end, we perform the Task Packing "in place", taking advantage of idle cycles generated at synchronization points of unbalanced applications. Evaluations are carried out on a heterogeneous platform using FT and miniFE benchmarks. Results showed that our proposal generates low overhead. In addition the amount of freed CPUs are related to a load imbalance metric which can be used as a prediction for it.

show abstract

Scheduling parallel jobs on multicore clusters using CPU oversubscription

Cited by 11 publications

References 21 publications

Parallel Computing with P2P Desktop Grids

Parallel Computing with P2P Desktop Grids

Checkpoint/restart approaches for a thread-based MPI runtime

Task Packing: Efficient task scheduling in unbalanced parallel programs to maximize CPU utilization

Contact Info

Product

Resources

About