Presentation is loading. Please wait.

Presentation is loading. Please wait.

Multi-Resource Packing for Cluster Schedulers Robert Grandl Aditya Akella Srikanth Kandula Ganesh Ananthanarayanan Sriram Rao.

Similar presentations


Presentation on theme: "Multi-Resource Packing for Cluster Schedulers Robert Grandl Aditya Akella Srikanth Kandula Ganesh Ananthanarayanan Sriram Rao."— Presentation transcript:

1 Multi-Resource Packing for Cluster Schedulers Robert Grandl Aditya Akella Srikanth Kandula Ganesh Ananthanarayanan Sriram Rao

2 Diverse Resource Requirements Tasks need varying amounts of each resource  E.g., Memory [100MB to 17GB] CPU [2% of a core to 6 cores] Need to match tasks with machines based on resource Demands for resources are not correlated  Correlation coefficient across resource demands  [–0.11, 0.33]

3 Current Schedulers do not Pack Resources allocated in terms of “slots” Resource Fragmentation T1: 2 GB T2: 2 GB T3: 4 GB 4 GB Memory Machine A 4 GB Memory Machine B Current SchedulersPacker Schedulers T1: 2 GB T2: 2 GB T3: 4 GB 4 GB Memory Machine A 4 GB Memory Machine B

4 Current Schedulers do not Pack Resources allocated in terms of “slots” Over-allocation 20 MB/s In Nw. T1: 2 GB Mem Current Schedulers 4 GB Memory 20 MB/s In Nw. Machine A 20 MB/s In Nw. T2: 2 GB Mem 20 MB/s In Nw. 20 MB/s In Nw. Packer Schedulers 20 MB/s In Nw. T1: 2 GB Mem 4 GB Memory 20 MB/s In Nw. Machine A 20 MB/s In Nw. T2: 2 GB Mem T3: 2 GB Mem

5 Current Schedulers do not Pack Slots allocated purely on fairness considerations Cluster [18 Cores, 36 GB] / Job: [Task Prof.], # tasks A[1 Core, 2 GB], 18 B[3 Cores, 1 GB], 6 C 6 tasks 2 tasks 18 cores 16 GB 6 tasks 2 tasks 6 tasks 2 tasks 18 cores 16 GB 18 cores 16 GB A B C t 2t 3t 18 tasks 0 tasks 18 cores 36 GB 6 tasks 0 tasks6 tasks 18 cores 6 GB 18 cores 6 GB A B C t 2t 3t Durations: DRF Packer Resources used: DRF share = 1/3Resources used: Packer Current Schedulers Packer Schedulers A: 3t B: 3t C: 3t Durations: A: t B: 2t C: 3t 33% improvement

6 It is all about packing ? Multi-dimensional bin packing is NP-hard for #dimens. ≥ 2  Several heuristics proposed  But they do not apply here … size of the ball, contiguity of allocation, resource demands are elastic in time Will perfect packing suffice ? Competing objectives: Cluster utilization vs. Job completion times vs. Fairness

7 Something reasonably simple and which can be applied Intuition behind the solution Cluster efficiencyJob completion time Cluster efficiency Fairness

8 Tetris Pack tasks along multiple resources  Cosine similarity between task demand vector and machine resource vector Multi-resource version of SRTF  Favor jobs with small remaining duration and small resource consumption Incorporate Fairness  Fairness knob  (0, 1] f → 0 close to perfect fairness f = 1 most efficient scheduling A T F 1: while (resources R are free) 2: among  FJ  jobs furthest from fair share 3: score (j) = 4: max task t in j, demand(t) ≤ R A(t, R) +  T(j) 5: pick j*, t* = argmax score(j) 6: R = R – demand(t*) 7: end while (simplified) Scheduling procedure

9 Learning task requirements  From tasks that have finished in the same phase  Coefficient of variation  [0.022, 0.41]  Collecting statistics from recurring jobs Task Requirements and resource usages Resource Tracker  measure actual usage of resources  enforce allocations  aware of activities on the cluster other than tasks assignment: ingest and evacuation  Peak usage demands estimates for tasks Machine - In Network 850 1024 0 512 MBytes / sec Time (sec) In Network Used In Network Free Task In Network Estimates Resource Tracker

10 Prototype atop Hadoop 2.3 Evaluation  Tetris as a pluggable scheduler to RM  Implement RT as a NM service  Modified AM/RM resource allocation protocol Cluster capacity: 250 Nodes 4 hour synthetic workload 60 jobs with complementary task demands Reduction (%) in Job Duration - DRF CDF Reduces average job duration by up to 40% Reduces makespan by 39% Large scale evaluation

11 Evaluation Facebook production traces analysis Fairness knob: fewer than 6% of jobs slow down; by not more than 8% on average Knob value of 0.75 offers nearly the best possible efficiency with little unfairness Trace-driven simulation Fairness Knob - Fair Slowdown (%) Fairness Knob - DRF Slowdown (%)

12 Conclusion Identify the importance of scheduling all relevant resources in a cluster Resource Fragmentation Over-allocation and Interference New scheduler that pack tasks along multiple resources Reduce makespan Job Completion Time Enable a trade-off between packing efficiency and fairness Fairness Knob Come and see our poster !


Download ppt "Multi-Resource Packing for Cluster Schedulers Robert Grandl Aditya Akella Srikanth Kandula Ganesh Ananthanarayanan Sriram Rao."

Similar presentations


Ads by Google