Approaches for improving tuple space search-based table lookup

Abstract

Listen

Translate article

One way for improving the tuple space search-based table lookup in the software system (e.g., flow matching process of Open vSwitch) is to reduce the number of tuple spaces to be examined. For this purpose, we discuss two approaches based on space-efficient data structures, Bloom filter and simTable proposed in this paper. Click-based experiments show that Bloom filter and simTable noticeably reduce the number of tuple spaces to be examined (i.e., the number of tuple spaces to be examined is 1 in most cases even with 121 tuple spaces). However, two approaches show different results in terms of the table lookup time. Bloom filter without hardware parallelism causes noticeable processing overhead that is heavily affected by the number of tuple spaces while simTable causes low processing overhead regardless of the number of tuple spaces. As a result, the use of Bloom filter/simTable increases/reduces the table lookup time compared to the original tuple space search-based table lookup (e.g., from 1917ns to 5032ns/432ns in case of 25 tuple spaces).