Expressing pipeline parallelism using TBB constructs

Abstract

Task-based libraries such as Intel's Threading Building Blocks (TBB) provide higher levels of abstraction than threads for parallel programming. Work remains, however, to determine how straightforward it is to use these libraries to express various patterns of parallelism. This case study focuses on a particular pattern: pipeline parallelism. We attempted to transform three representative applicationscontent-based image retrieval, compression and video encoding -to pipelines using TBB. We successfully converted two of the three applications. In the successful cases we discuss our transformation process and contrast the expressivity and performance of our implementations to existing Pthreads versions; in the unsuccessful case, we detail what the challenges were and propose potential solutions.

Deterministic scale-free pipeline parallelism with hyperqueues

“…Figure 11 shows speedup for dedup in the pthreads, TBB and Swan programming models. While Reed et al demonstrated improved performance of their TBB implementation relative to the pthreads implementation in PARSEC 2.1 [22], our evaluation using PARSEC 3.0 shows that the TBB implementation is slower than the pthreads implementation. The Swan implementation with hyperqueues outperforms the pthread version by at least 12% and up to 30% in the region of 6-8 threads.…”

“…However, it requires thoroughly restructuring the input stage in such a way that it can be called repeatedly to produce the next file [22]. To this end, its internal state must be made explicit (i.e., its current position in the traversal of the directory tree) and passed as an argument to the first stage.…”

“…Reed et al observed that dedup exhibits a nested pipeline [22]. The outer pipeline, handling large chunks, consists of three stages: Fragment, InnerPipeline and Output.…”