Job scheduling with adjusted runtime estimates on production supercomputers

Tang, Wei; Desai, Narayan; Buettner, Daniel; Lan, Zhiling

doi:10.1016/j.jpdc.2013.02.006

Cited by 25 publications

(12 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Then based on the size of these slots and each job's size, our design will use B&B or Greedy algorithm to make the following scheduling decision. First, it puts job C into the first slot(assign nodes 1, 2, 3 to C); then put job A into the second slot (6,7,8,9,10,11 to A); 15, 16, 17, 18 to B; 19, 20 to E. Apparently, our design can guarantee that every job gets a compact allocation while maintain high system utilization. list, the grey ones are been occupied by current running jobs.…”

Section: An Examplementioning

confidence: 98%

“…Many studies seek to improve the performance of this classic scheduling paradigm. Tang et al made refinement about user's estimated job runtime in order to make the backfilling more efficient [6] [8]. They also designed a walltime-aware job allocation strategy, which adjacently packs jobs that finish around the same time, in order to minimize resource fragmentation caused by job length discrepancy [7].…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Balancing job performance with system performance via locality-aware scheduling on torus-connected systems

Zhou

Tang

et al. 2014

2014 IEEE International Conference on Cluster Computing (CLUSTER)

Self Cite

View full text Add to dashboard Cite

Torus-connected network is widely used in modern supercomputers due to its linear per node cost scaling and its competitive overall performance. Job scheduling system plays a critical role for the efficient use of supercomputers. As supercomputers continue growing in size, a fundamental problem arises: how to effectively balance job performance with system performance on torus-connected machines? In this work, we will present a new scheduling design named window-based locality-aware scheduling. Our design contains three novel features. First, rather than oneby-one job scheduling, our design takes a "window" of jobs, i.e. multiple jobs, into consideration for job prioritizing and resource allocation. Second, our design maintains a list of slots to preserve node contiguity information for resource allocation. Finally, we formulate our scheduling decision making into a 0-1 Multiple Knapsack Problem and present two algorithms to solve the problem. A series of trace-based simulations using job logs collected from production supercomputers indicate that this new scheduling design has real potentials and can effectively balance job performance and system performance.

show abstract

Section: An Examplementioning

confidence: 98%

Section: Related Workmentioning

confidence: 99%

Balancing job performance with system performance via locality-aware scheduling on torus-connected systems

Zhou

Tang

et al. 2014

2014 IEEE International Conference on Cluster Computing (CLUSTER)

Self Cite

View full text Add to dashboard Cite

show abstract

“…They can be very accurate, but building good models is labor-intensive and hard to automate. Moreover, user estimates of application runtimes are often highly inaccurate [1]. Code analysis automates this process, but it usually restricts itself to coarse-grained decisions such as the choice of the best acceleration device for optimizing performance [6].…”

Section: Related Workmentioning

confidence: 99%

“…We do not specify the network capacity between provisioned machine instances in our manifests, as current clouds do not allow a user to specify such properties 1 . A logical extension of this work would be to also specify available bandwidth between resources, provided that the underlying cloud can take such requests into account.…”

Section: Handling Arbitrary Applicationsmentioning

confidence: 99%

“…This allows them in principle to use the exact type and number of resources their application needs. However, this flexibility also comes as a curse as choosing the best resource configuration for an application becomes extremely difficult: cloud users often find it very hard to accurately estimate the requirements of complex applications [1].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Heterogeneous Resource Selection for Arbitrary HPC Applications in the Cloud

Iordache¹,

Buyukkaya²,

Pierre³

2015

Distributed Applications and Interoperable Systems

View full text Add to dashboard Cite

Abstract. Cloud infrastructures offer a wide variety of resources to choose from. However, most cloud users ignore the potential benefits of dynamically choosing cloud resources among a wide variety of VM instance types with different configuration/cost tradeoffs. We propose to automate the choice of resources that should be assigned to arbitrary non-interactive applications. During the first executions of the application, the system tries various resource configurations and builds a custom performance model for this application. Thereafter, cloud users can specify their execution time or financial cost constraints, and let the system automatically select the resources which best satisfy this constraint.

show abstract

Qespera: an adaptive framework for prediction of queue waiting times in supercomputer systems

Murali

Vadhiyar

2015

Concurrency and Computation

View full text Add to dashboard Cite

Production parallel systems are space-shared, and resource allocation on such systems is usually performed using a batch queue scheduler. Jobs submitted to the batch queue experience a variable delay before the requested resources are granted. Predicting this delay can assist users in planning experiment time-frames and choosing sites with less turnaround times and can also help meta-schedulers make scheduling decisions. In this paper, we present an integrated adaptive framework, Qespera, for prediction of queue waiting times on parallel systems. We propose a novel algorithm based on spatial clustering for predictions using history of job submissions and executions. The framework uses adaptive set of strategies for choosing either distributions or summary of features to represent the system state and to compare with history jobs, varying the weights associated with the features for each job prediction, and selecting a particular algorithm dynamically for performing the prediction depending on the characteristics of the target and history jobs. Our experiments with real workload traces from different production systems demonstrate up to 22% reduction in average absolute error and up to 56% reduction in percentage prediction error over existing techniques. We also report prediction errors of less than 1 h for a majority of the jobs. P. MURALI AND S. VADHIYAR Figure 1. Queue waiting times seen in production HPC systems.On space-sharing systems, it is expected that users request a set of compute nodes for a particular duration of time. With multiple users contending for the compute resources, a batch queue submission incurs time because of waiting in the queue before the resources necessary for its execution are allocated. The queue waiting time ranges from a few seconds to even a few days on most production systems. Figure 1 shows the box plot of queue waiting times observed in the workloads at five production systems featured in parallel workloads archive (PWA) [5]. In each box, the lower and upper edges denote the lower and upper quartile of the set of queue waiting times and the middle line shows the median. The whiskers mark the interquartile deviation (1.5 ) from the median. From the large number of outlier points with high waiting times and the small values of lower quartile and median, we can see that queue waiting times vary widely within and across systems.Prediction of queue waiting times for jobs enables users to estimate job turnaround time, which can be used for planning and task management. Additionally, reliable predictions will allow the users to tune job parameters like requested number of nodes or estimated running time to achieve faster response times. Such predictions can also be efficiently used by a meta-scheduler to make automatic scheduling decisions for selecting the appropriate number of processors and queues for job execution. Usage statistics of predictors like queue bounds estimation from time series (QBETS) [6] that was used to provide bounds on the queue waiting times in TeraGrid [7] a...

show abstract

Job scheduling with adjusted runtime estimates on production supercomputers

Cited by 25 publications

References 27 publications

Balancing job performance with system performance via locality-aware scheduling on torus-connected systems

Balancing job performance with system performance via locality-aware scheduling on torus-connected systems

Heterogeneous Resource Selection for Arbitrary HPC Applications in the Cloud

Qespera: an adaptive framework for prediction of queue waiting times in supercomputer systems

Contact Info

Product

Resources

About