In last decade, data analytics have rapidly progressed from traditional disk-based processing tomodern in-memory processing. However, little effort has been devoted at enhancing performance at micro-architecture level. This paper characterizes the performance of in-memory data analytics using Apache Spark framework. We use a single node NUMA machine and identify the bottlenecks hampering the scalability of workloads. We also quantify the inefficiencies at micro-architecture level for various data analysis workloads. Through empirical evaluation, we show that spark workloads do not scale linearly beyond twelve threads, due to work time inflation and thread level load imbalance. Further, at the micro-architecture level, we observe memory bound latency to be the major cause of work time inflation.

Sheer increase in volume of data over the last decade has triggered research in cluster computing frameworks that enable web enterprises to extract big insights from big data. While Apache Spark is gaining popularity for exhibiting superior scale-out performance on the commodity machines, the impact of data volume on the performance of Spark based data analytics in scale-up configuration is not well understood. We present a deep-dive analysis of Spark based applications on a large scale-up server machine. Our analysis reveals that Spark based data analytics are DRAM bound and do not benefit by using more than 12 cores for an executor. By enlarging input data size, application performance degrades significantly due to substantial increase in wait time during I/O operations and garbage collection, despite 10 % better instruction retirement rate (due to lower L1 cache misses and higher core utilization). We match memory behaviour with the garbage collector to improve performance of applications between 1.6x to 3x.

While cluster computing frameworks are contin-uously evolving to provide real-time data analysis capabilities,Apache Spark has managed to be at the forefront of big data an-alytics for being a unified framework for both, batch and streamdata processing. However, recent studies on micro-architecturalcharacterization of in-memory data analytics are limited to onlybatch processing workloads. We compare micro-architectural per-formance of batch processing and stream processing workloadsin Apache Spark using hardware performance counters on a dualsocket server. In our evaluation experiments, we have found thatbatch processing are stream processing workloads have similarmicro-architectural characteristics are bounded by the latency offrequent data access to DRAM. For data accesses we have foundthat simultaneous multi-threading is effective in hiding the datalatencies. We have also observed that (i) data locality on NUMAnodes can improve the performance by 10% on average and(ii)disabling next-line L1-D prefetchers can reduce the executiontime by up-to 14% and (iii) multiple small executors can provideup-to 36% speedup over single large executor