Vectorized Bloom filters for advanced SIMD processors

Abstract

Analytics are at the core of many business intelligence tasks. Efficient query execution is facilitated by advanced hardware features, such as multi-core parallelism, shared-nothing low-latency caches, and SIMD vector instructions. Only recently, the SIMD capabilities of mainstream hardware have been augmented with wider vectors and non-contiguous loads termed gathers. While analytical DBMSs minimize the use of indexes in favor of scans based on sequential memory accesses, some data structures remain crucial. The Bloom filter, one such example, is the most efficient structure for filtering tuples based on their existence in a set and its performance is critical when joining tables with vastly different cardinalities. We introduce a vectorized implementation for probing Bloom filters based on gathers that eliminates conditional control flow and is independent of the SIMD length. Our techniques are generic and can be reused for accelerating other database operations. Our evaluation indicates a significant performance improvement over scalar code that can exceed 3X when the Bloom filter is cache-resident.

SIMD compression and the intersection of sorted integers

“…For example, our implementation of SIMD GALLOPING relies on a scalar binary search like the conventional galloping. However, recent Intel processors have introduced gather instructions (vpgatherdd) that can retrieve at once several integers from various locations . Such instructions could accelerate binary search.…”

ViViD Cuckoo Hash: Fast Cuckoo Table Building in SIMD

“…With ViViD Cuckoo we could address the two main issues that caused the build to become a bottleneck for the Cuckoo Join: the chained unpredictable control dependencies and the data dependencies between cuckoo paths. We extended the methods described by [Polychroniou et al 2015] and [Polychroniou and Ross 2014] to use new capabilities of AVX-512 and to take advantage of logical operations and the dispersion provided by the hash functions.…”

“…The left side receives all the keys that will be replaced for new ones from the relation, in the example the keys 675 and 76. This process is based on [Polychroniou and Ross 2014] approach using a permutation table kept in memory to speed the process.…”

SIMD-accelerated regular expression matching

“…In databases, however, regular expressions are used as a boolean filter and the matching can skip large portions of each input string, making lockstep processing wasteful. Vectorization using data-parallel processing of multiple input instances was used to accelerate database operators on CPUs and Xeon Phi co-processors [8,9]. Our design not only traverses the DFA for multiple strings in parallel, but also accesses the strings at arbitrary offsets rather than in lockstep, buffers multiple bytes per access, and replaces strings as soon as they are accepted or rejected by the DFA, in order to fully utilize the SIMD lanes.…”