Symbiotic Space-Sharing on SDSC’s DataStar System

Weinberg, Jonathan; Snavely, Allan

doi:10.1007/978-3-540-71035-6_10

Cited by 17 publications

(19 citation statements)

References 28 publications

(26 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We also found that pairing communication-bound applications may yield acceptable slowdown [15] [20]. While our work focused on computation and communication performance, the study in [18] focused on memory and I/O, confirming the benefits from coscheduling. However, there are also cases where an application can exploit the performance potential of all cores per node well and space sharing may perform better.…”

Section: Introductionsupporting

confidence: 62%

“…In earlier work, we have shown that better results may be obtained if rather using the additional cores for other applications with complementary characteristics [14][16] [20], as also found by other researchers [18]. We call such resource allocation semi time sharing since the cores are partitioned among the applications, i.e.…”

Section: Introductionmentioning

confidence: 85%

“…In regards to semi time sharing, studies found complementary characteristics of the coscheduled jobs to perform better due to balancing the usage of system resources [16] [18]. Spreading out jobs to different nodes and semi time sharing the resources per node among multiple jobs may perform better than dedicated allocation of all node resources to one job [14][15] [18].…”

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

“…Spreading out jobs to different nodes and semi time sharing the resources per node among multiple jobs may perform better than dedicated allocation of all node resources to one job [14][15] [18]. The experiments in [18] were carried out with only 4 nodes and therefore limited communication but the experiments in [14][15] on up to 64 nodes show that the benefits of semi time sharing vs. dedicated allocation scale to larger number of nodes. Characteristics which were found important for job combinations are whether the job is CPUbound, cache-bound, memory-bound, network-bound, or disk-bound [16][18] but also which access patterns are applied [10] [20].…”

Section: Related Workmentioning

confidence: 99%

See 4 more Smart Citations

Job Scheduling with Lookahead Group Matchmaking for Time/Space Sharing on Multi-core Parallel Machines

Zeng

Sodan

2009

Job Scheduling Strategies for Parallel Processing

View full text Add to dashboard Cite

Abstract. Multi-core nodes of parallel machines may only provide gradual performance improvement per application due to competition on resources like the cache. As shown in our earlier work, spreading out applications over as many nodes as possible or letting different applications with potentially complementary characteristics (semi time) share each node by allocating different cores to them may provide better performance. In the latter case, groups of jobs may be necessary to obtain balanced resource utilization due to different sizes of jobs. We present a scheduler G-LOMARC-TS which can match groups of jobs and consider both space-and time-sharing allocation. Since matchmaking may select jobs further down in the waiting queue, fairness in regards to possible delays of the other jobs is watched and delays are kept within certain bounds. This results in a large number of possible combinations. A number of heuristics to select the most promising combinations make it possible to deal with the NP-completeness of the problem. We show that our scheduler improves utilization of high-load phases by about 27% and subsequently average response times by about 36% (and 53% for long jobs) compared to space sharing scheduling for normal workloads. Additionally the scheduler can handle much higher workloads than a space-sharing scheduler.

show abstract

Section: Introductionsupporting

confidence: 62%

Section: Introductionmentioning

confidence: 85%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

See 3 more Smart Citations

Job Scheduling with Lookahead Group Matchmaking for Time/Space Sharing on Multi-core Parallel Machines

Zeng

Sodan

2009

Job Scheduling Strategies for Parallel Processing

View full text Add to dashboard Cite

show abstract

The case for colocation of high performance computing workloads

Breslow

Porter

Tiwari

et al. 2013

Concurrency and Computation

View full text Add to dashboard Cite

The current state of practice in supercomputer resource allocation places jobs from different users on disjoint nodes both in terms of time and space. While this approach largely guarantees that jobs from different users do not degrade one another's performance, it does so at high cost to system throughput and energy efficiency. This focused study presents job striping, a technique that significantly increases performance over the current allocation mechanism by colocating pairs of jobs from different users on a shared set of nodes. To evaluate the potential of job striping in large-scale environments, the experiments are run at the scale of 128 nodes on the state-of-the-art Gordon supercomputer. Across all pairings of 1024 process network-attached storage parallel benchmarks, job striping increases mean throughput by 26% and mean energy efficiency by 22%. On pairings of the real applications Gyrokinetic Toroidal Code (GTC), Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS), and MIMD Lattice Computation (MILC) at equal scale, job striping improves average throughput by 12% and mean energy efficiency by 11%. In addition, the study provides a simple set of heuristics for avoiding low performing application pairs. 239 Figure 3. Increase in system throughput (STP) over compact when applying job spreading and striping to the NAS parallel benchmarks and GTC, LAMMPS, and MILC. Figure 3 shows the performance results for the first set of experiments with the NPBs and the second set of experiments with GTC, LAMMPS, and MILC. For the NPBs, the mean performance increase from job spreading is 50%. If one examines striped coschedules of non-identical NPBs, the average performance increase is 26%. If one selects the best running mate other than embarrassingly parallel (EP) for each benchmark, then the average increase in performance is 36%. We choose to exclude EP because EP is minimally contentious. Each EP task's working set fits entirely in the private levels of cache, and EP spends very little time in active communication. Because of these traits, EP universally causes every application that it stripes with to achieve its best striped performance. Thus, for the sake of fairness and realism, we exclude these results from the 'Best' average. For the NPBs, random striping yields about 50% of the performance benefit of job spreading, and striping each job with its best running mate provides 70% of the performance benefit of spreading. This trend continues for real applications as well. For GTC, LAMMPS, and MILC, job spreading increases throughput by 23% and the mean heterogeneous striping and the mean best striping improve performance by 12% and 16%, respectively. PERFORMANCE RESULTS Compact versus spreading versus striping Network-attached storage parallel benchmarksIn this section, we examine the increase in collective throughput and energy efficiency for pairs of striped NPBs. The results are presented in Figure 4. For completeness, we run all pairwise combinations. This includes both heterogeneous pairings ...

show abstract