Parallel Bloom Filter on Xeon Phi Many-Core Processors

Ni, Sheng Huoo; Guo, Rentong; Liao, Xiaofei; Jin, Hai

doi:10.1007/978-3-319-27122-4_27

Cited by 2 publications

(2 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We present the algorithm used, the processor, and the maximum performance for IPv4 and IPv6. Finally, we also shown if the performance collected considered the PCIe data transfers, which are not necessary on our algorithm when using the KNL A Parallel Bloom Filter (PBF) was implemented in the Intel Phi [19]. PBF was proposed to reduce synchronization overhead and improve cache locality.…”

Section: Tablementioning

confidence: 99%

See 1 more Smart Citation

High-performance IP lookup using Intel Xeon Phi: a Bloom filters based approach

Lucchesi

Drummond

Teodoro

2018

J Internet Serv Appl

View full text Add to dashboard Cite

IP lookup is a core operation in packet forwarding, which is implemented using a Longest Prefix Matching (LPM) algorithm to find the next hop for an input address. This work proposes and evaluates the use of parallel processors to deploy an optimized IP lookup algorithm based on Bloom filters. We target the implementation on the Intel Xeon Phi (Intel Phi) many-core coprocessor and on multi-core CPUs, and also evaluate the cooperative execution using both computing devices with several optimizations. The experimental evaluation shows that we were able to attain high IP lookup throughputs of up to 182.7 Mlps (Million packets per second) for IPv6 packets on a single Intel Phi. This performance indicates that the Intel Phi is a very promising platform for deployment of IP lookup. We also compared the Bloom filters to an efficient approach based on the Multi-Index Hybrid Trie (MIHT) in which the Bloom filters was up to 5.1× faster. We also propose and evaluate the cooperative use of CPU and Intel Phi in the IP lookup, which resulted in an improvement of about 1.3× as compared to the execution using only the Intel Phi.

show abstract

Section: Tablementioning

confidence: 99%

“…The ILP optimization focused on the hashing calculations because it is the most compute intensive stage of the algorithm. The original work [27] and previous implementations of algorithms employing Bloom filters to the LPM problem [18,19] do not discuss their decisions and reasons on the hash functions used.…”

Section: Instruction-level Parallelism (Ilp)mentioning

confidence: 99%