source: docs/PACT2011/08-conclusions.tex @ 1068

Last change on this file since 1068 was 1023, checked in by cameron, 9 years ago

Updated main with conclusions

File size: 1.7 KB
3This paper has examined energy efficiency and performance
4characteristics of four XML parsers considered over three
5generations of Intel processor architecture and shown that
6parsers based on parallel bit stream technology have dramatically
7better performance, energy efficiency and scalability than
8traditional byte-at-a-time parsers widely deployed in current
9software.  Based on a novel application of the short vector
10SIMD technology commonly found in commodity processors of
11all kinds, parallel bit stream technology scales well with
12improvements in processor SIMD capabilities.  With the recent
13introduction of the first generation of Intel processors that
14incorporate AVX technology, the change to 3-operand
15form SIMD operations has delivered a substantial benefit
16for the Parabix2 parsers simply through recompilation.
17Restructuring of Parabix2 to take advantage of the 256-bit SIMD
18capabilities also delivered a substantial reduction in
19instruction count, but without corresponding performance
20benefits in the first generation of AVX implementations.
22There are many directions for further research.   These
23include compiler and tools technology to automate the low-level
24programming tasks inherent in building parallel bit stream
25applications, widening the research by applying the techniques
26to other forms of text analysis and parsing, and further
27investigation of the interaction between parallel bit
28stream technology and processor architecture.  Two promising
29avenues include investigation of GPGPU approaches to parallel
30bit stream technology and the leveraging of the intraregister parallelism
31inherent in this approach to also take advantage of the intrachip
32parallelism of multicore processors.
Note: See TracBrowser for help on using the repository browser.