Main-memory scan sharing for multi-core CPUs

Abstract

Computer architectures are increasingly based on multi-core CPUs and large memories. Memory bandwidth, which has not kept pace with the increasing number of cores, has become the primary processing bottleneck, replacing disk I/O as the limiting factor. To address this challenge, we provide novel algorithms for increasing the throughput of Business Intelligence (BI) queries, as well as for ensuring fairness and avoiding starvation among a concurrent set of such queries. To maximize throughput, we propose a novel FullSharing scheme that allows all concurrent queries, when performing base-table I/O, to share the cache belonging to a given core. We then generalize this approach to a BatchSharing scheme that avoids thrashing on "agg-tables"-hash tables that are used for aggregation processing-caused by execution of too many queries on a core. This scheme partitions queries into batches such that the working-set of agg-table entries for each batch can fit into a cache; an efficient sampling technique is used to estimate selectivities and working-set sizes for purposes of query partitioning. Finally, we use lottery-scheduling techniques to ensure fairness and impose a hard upper bound on staging time to avoid starvation. On our 8-core testbed, we were able to completely remove the memory I/O bottleneck, increasing throughput by a factor of 2 to 2.5, while also maintaining fairness and avoiding starvation.

Predictable performance and high query concurrency for data analytics

“…There may be opportunities to optimize this final stage, e.g., by sharing work among aggregation operators, depending on the current query mix. This optimization is orthogonal to Cjoin and can employ existing techniques [10,24].…”

“…Cooperative scans [31] improve data sharing across concurrent scans by dynamically scheduling queries and their data requests, taking into account current system conditions. Qiao et al [24] have investigated shared memory scans as a specific form of work sharing in multi-core systems: by scheduling concurrent queries carefully, tuple accesses can be coordinated in the processor's cache. Another possibility is to recycle intermediate results and share them with subsequent queries [17].…”

“…Recent work [10,24] has investigated work-sharing techniques for the computation of aggregates on chip multiprocessors. The developed techniques essentially synchronize the execution of different aggregation operators in order to reduce contention on the hash tables used for aggregate computation.…”

High availability, elasticity, and strong consistency for massively parallel scans over relational data

“…Shared scans have been implemented in the context of disk-based systems such as Red Brick DW [34], DB2 UDB [57], and MonetDB/X100 [79]. Shortly before Crescando, Raman et al [63,64] had demonstrated that shared scans in main memory can provide predictable, low ("constant") access latency.…”

Scaling up analytical queries with column-stores

“…Circular scans are implemented in a variety of systems [7,8,9,14] and they can sustain a large number of concurrent tablescans by reducing contention for the buffer pool and the number of accesses to the underlying storage devices. More elaborate techniques for shared scans with coordination are developed for column-stores [29] and for main-memory shared scans [21].…”