Speculative Out-Of-Order Event Processing with Software Transactional Memory
Research output: Contribution to conferences › Paper › Contributed
Contributors
Abstract
In event stream applications, events flow through a net-
work of components that perform various types of opera-
tions, e.g., filtering, aggregation, transformation. When the
operation only depends on the input events, one can triv-
ially parallelize its processing by replicating the associated
components. This is not possible, however, with stateful
components or when there exist dependencies between the
events. Parallel versions of a number of simple stream min-
ing operators have been designed, but, in general, complex
and user-defined operators are limited by single thread per-
formance. In this paper, we propose leveraging the pro-
cessing capabilities of multi-core processors to improve the
efficiency of stateful components using optimistic paralleliza-
tion techniques (as provided by transactional memory). We
show that, even though some speculative event executions
might need to be disregarded, the overall throughput in-
creases noticeably in the general case and latency can be
reduced by pre-processing out-of-order events. Moreover,
we show how simple conflict predictors can boost the paral-
lelism even more and reduce the amount of resources used
for a given level of parallelism.
work of components that perform various types of opera-
tions, e.g., filtering, aggregation, transformation. When the
operation only depends on the input events, one can triv-
ially parallelize its processing by replicating the associated
components. This is not possible, however, with stateful
components or when there exist dependencies between the
events. Parallel versions of a number of simple stream min-
ing operators have been designed, but, in general, complex
and user-defined operators are limited by single thread per-
formance. In this paper, we propose leveraging the pro-
cessing capabilities of multi-core processors to improve the
efficiency of stateful components using optimistic paralleliza-
tion techniques (as provided by transactional memory). We
show that, even though some speculative event executions
might need to be disregarded, the overall throughput in-
creases noticeably in the general case and latency can be
reduced by pre-processing out-of-order events. Moreover,
we show how simple conflict predictors can boost the paral-
lelism even more and reduce the amount of resources used
for a given level of parallelism.
Details
Original language | English |
---|---|
Pages | 265-275 |
Number of pages | 11 |
Publication status | Published - 2008 |
Peer-reviewed | No |
Conference
Title | DEBS '08 - the second international conference on Distributed event-based systems, ACM, 2008. |
---|---|
Abbreviated title | DEBS '08 |
Conference number | |
Duration | 1 April 2008 |
Degree of recognition | International event |
Location | |
City | Rom |
Country | Italy |
Keywords
Research priority areas of TU Dresden
DFG Classification of Subject Areas according to Review Boards
Keywords
- Event stream processing, Software transactional memory