EPiC Series in Computing Programming by Composing Filters
release_iocbvtfodnciplgzphwt34swgy
by
Jeffrey Fischer, Rupak Majumdar
2017 Volume 46
Abstract
We present a formal model for event-processing pipelines. Event-processing pipelines appear in a large number of domains, from control of cyber-physical systems (CPS), to large scale data analysis, to Internet-of-things applications. These applications are characterized by stateful transformations of event streams, for example, for the purposes of sensing, computation, and actuation of inner control loops in CPS applications, and for data cleaning, analysis, training, and querying in data analytics applications. Our formal model provides two abstractions: streams of data, and stateful, probabilistic, filters, which transform input streams to output streams probabilistically. Programs are compositions of filters. The filters are scheduled and run by an explicit, asynchronous, scheduler. We provide a transition system semantics for such programs based on infinite-state Markov decision processes. We characterize when a program is scheduler-independent, that is, provides the same observable behavior under every scheduler, based on local commutativity.
In text/plain
format
Archived Files and Locations
application/pdf 459.2 kB
file_naiae45mp5hhnotvngqg7cvh3m
|
web.archive.org (webarchive) easychair.org (web) |
article-journal
Stage
unknown
Year 2017
access all versions, variants, and formats of this works (eg, pre-prints)