Moving Hadoop into the Cloud with Flexible Slot Management and Speculative Execution

Load imbalance is a major source of overhead in parallel programs such as MapReduce. Due to the uneven distribution of input data, tasks with more data become stragglers and delay the overall job completion. Running Hadoop in a private cloud opens up opportunities for expediting stragglers with more resources but also introduces problems that often outweigh the performance gain: (1) performance interference from co-running jobs may create new stragglers; (2) there exists a semantic gap between the Hadoop task management and resource pool-based virtual cluster management preventing tasks from using resources efficiently. In this paper, we strive to make Hadoop more resilient to data skew and more efficient in cloud environments. We present <italic>FlexSlot</italic>, a user-transparent task slot management scheme that automatically identifies map stragglers and resizes their slots accordingly to accelerate task execution. FlexSlot adaptively changes the number of slots on each virtual node to balance the resource usage so that the pool of resources can be efficiently utilized. FlexSlot further improves mitigation of data skew with an adaptive speculative execution strategy. Experimental results show that FlexSlot effectively reduces job completion time up to <inline-formula><tex-math notation="LaTeX">$47.2$</tex-math><alternatives> <inline-graphic xlink:href="guo-ieq1-2587641.gif"/></alternatives></inline-formula> percent compared to stock Hadoop and two recently proposed skew mitigation and speculative execution approaches.

[1]  Hai Jin,et al.  LEEN: Locality/Fairness-Aware Key Partitioning for MapReduce in the Cloud , 2010, 2010 IEEE Second International Conference on Cloud Computing Technology and Science.

[2]  Ajay Gulati.  VMware distributed resource Management : design , Implementation , and lessons learned , 2022 .

[3]  Xiaobo Zhou,et al.  iShuffle: Improving Hadoop Performance with Shuffle-on-Write , 2017, IEEE Transactions on Parallel and Distributed Systems.

[4]  Y. Charlie Hu,et al.  PIKACHU: How to Rebalance Load in Optimizing MapReduce On Heterogeneous Clusters , 2013, USENIX Annual Technical Conference.

[5]  Beng Chin Ooi,et al.  The performance of MapReduce , 2010, Proc. VLDB Endow..

[6]  Roy H. Campbell,et al.  ARIA: automatic resource inference and allocation for mapreduce environments , 2011, ICAC '11.

[7]  Peter J. Varman,et al.  Defragmenting the cloud using demand-based resource allocation , 2013, SIGMETRICS '13.

[8]  Magdalena Balazinska,et al.  Skew-resistant parallel processing of feature-extracting scientific user-defined functions , 2010, SoCC '10.

[9]  Antony I. T. Rowstron,et al.  Bridging the tenant-provider gap in cloud services , 2012, SoCC '12.

[10]  Aman Kansal,et al.  Q-clouds: managing performance interference effects for QoS-aware clouds , 2010, EuroSys '10.

[11]  Albert G. Greenberg,et al.  Reining in the Outliers in Map-Reduce Clusters using Mantri , 2010, OSDI.

[12]  Calton Pu,et al.  Understanding Performance Interference of I/O Workload in Virtualized Cloud Environments , 2010, 2010 IEEE 3rd International Conference on Cloud Computing.

[13]  Boon Thau Loo,et al.  AutoTune: Optimizing Execution Concurrency and Resource Usage in MapReduce Workflows , 2013, ICAC.

[14]  Zhen Xiao,et al.  Improving MapReduce Performance Using Smart Speculative Execution Strategy , 2014, IEEE Transactions on Computers.

[15]  Carlo Curino,et al.  Apache Hadoop YARN: yet another resource negotiator , 2013, SoCC.

[16]  Yonggang Hu,et al.  DynMR: dynamic MapReduce with ReduceTask interleaving and MapTask backfilling , 2014, EuroSys '14.

[17]  Roy H. Campbell,et al.  Resource Provisioning Framework for MapReduce Jobs with Performance Goals , 2011, Middleware.

[18]  Hai Jin,et al.  Maestro: Replica-Aware Map Scheduling for MapReduce , 2012, 2012 12th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (ccgrid 2012).

[19]  Randy H. Katz,et al.  Wrangler: Predictable and Faster Jobs using Fewer Resources , 2014, SoCC.

[20]  Scott Shenker,et al.  Usenix Association 10th Usenix Symposium on Networked Systems Design and Implementation (nsdi '13) 185 Effective Straggler Mitigation: Attack of the Clones , 2022 .

[21]  Xiaohui Gu,et al.  CloudScale: elastic resource scaling for multi-tenant cloud systems , 2011, SoCC.

[22]  Irene Finocchi,et al.  On data skewness, stragglers, and MapReduce progress indicators , 2015, SoCC.

[23]  Prashant J. Shenoy,et al.  A platform for scalable one-pass analytics using MapReduce , 2011, SIGMOD '11.

[24]  Jennifer L. Wong,et al.  To hardware prefetch or not to prefetch?: a virtualized environment study and core binding approach , 2013, ASPLOS '13.

[25]  M. Balazinska,et al.  An analysis of Hadoop usage in scientific workloads , 2013 .

[26]  T. N. Vijaykumar,et al.  Tarazu: optimizing MapReduce on heterogeneous clusters , 2012, ASPLOS XVII.

[27]  Changjun Jiang,et al.  FlexSlot: Moving Hadoop Into the Cloud with Flexible Slot Management , 2014, SC14: International Conference for High Performance Computing, Networking, Storage and Analysis.

[28]  Randy H. Katz,et al.  Improving MapReduce Performance in Heterogeneous Environments , 2008, OSDI.

[29]  Raghu Ramakrishnan,et al.  Sailfish: a framework for large scale data processing , 2012, SoCC '12.

[30]  Kun-Lung Wu,et al.  FLEX: A Slot Allocation Scheduling Optimizer for MapReduce Workloads , 2010, Middleware.

[31]  Yuan Yu,et al.  Dryad: distributed data-parallel programs from sequential building blocks , 2007, EuroSys '07.

[32]  Sanjay Ghemawat,et al.  MapReduce: Simplified Data Processing on Large Clusters , 2004, OSDI.

[33]  Magdalena Balazinska,et al.  SkewTune: mitigating skew in mapreduce applications , 2012, SIGMOD Conference.

[34]  Seyong Lee,et al.  PUMA: Purdue MapReduce Benchmarks Suite , 2012 .

[35]  Xiaobo Zhou,et al.  Improving MapReduce performance in heterogeneous environments with adaptive task tuning , 2014, Middleware.