No abstract

No abstract

Abstract. The MapReduce (MR) framework has become a standard tool for performing large batch computations-usually of aggregative nature-in parallel over a cluster of commodity machines. A significant share of typical MR jobs involves standard database-style queries, where it becomes cumbersome to specify map and reduce functions from scratch. To overcome this burden, higher-level languages such as HiveQL, PigLatin, and JAQL have been proposed to allow the automatic generation of MR jobs from declarative queries. We identify two major problems of these existing solutions: (i) they introduce new query languages and implement systems from scratch for the sole purpose of expressing MR jobs; and (ii) despite solving some of the major limitations of SQL, they still lack the flexibility required by big data applications. We propose BrackitMR, an approach based on the XQuery language with extended JSON support. XQuery not only is an established query language, but also has a more expressive data model and more powerful language constructs, enabling a much greater degree of flexibility. From a system design perspective, we extend an existing single-node query processor, Brackit, adding MR as a distributed coordination layer. Such heavy reuse of the standard query processor not only provides performance, but also allows for a more elegant design which transparently integrates MR processing into a generic query engine.

For decades, ARIES has been the standard for logging and recovery in database systems. ARIES offers important features like support for arbitrary workloads, fuzzy checkpoints, and transparent index recovery. Nevertheless, many modern inmemory database systems use more lightweight approaches that have less overhead and better multi-core scalability but only work well for the in-memory setting. Recently, a new class of high-performance storage engines has emerged, which exploit fast SSDs to achieve performance close to pure in-memory systems but also allow out-of-memory workloads. For these systems, ARIES is too slow whereas inmemory logging proposals are not applicable. In this work, we propose a new logging and recovery design that supports incremental and fuzzy checkpointing, index recovery, out-of-memory workloads, and low-latency transaction commits. Our continuous checkpointing algorithm guarantees bounded recovery time. Using per-thread logging with minimal synchronization, our implementation achieves near-linear scalability on multi-core CPUs. We implemented and evaluated these techniques in our LeanStore storage engine. For working sets that fit in main memory, we achieve performance close to that of an in-memory approach, even with logging, checkpointing, and dirty page writing enabled. For the out-of-memory scenario, we outperform a state-of-the-art ARIES implementation by a factor of two.

No abstract