TY - GEN

T1 - Reactors

T2 - 2018 ACM SIGMOD/PODS International Conference on Management of Data

AU - Shah, Vivek

AU - Vaz Salles, Marcos António

PY - 2018

Y1 - 2018

N2 - The requirements for OLTP database systems are becoming evermore demanding. Domains such as finance and computer gamesincreasingly mandate that developers be able to encode complexapplication logic and control transaction latencies in in-memorydatabases. At the same time, infrastructure engineers in these domainsneed to experiment with and deploy OLTP database architecturesthat ensure application scalability and maximize resource utilizationin modern machines. In this paper, we propose a relationalactor programming model for in-memory databases as a novel,holistic approach towards fulfilling these challenging requirements.Conceptually, relational actors, or reactors for short, are applicationdefined,isolated logical actors that encapsulate relations and processfunction calls asynchronously. Reactors ease reasoning aboutcorrectness by guaranteeing serializability of application-level functioncalls. In contrast to classic transactional models, however, reactorsallow developers to take advantage of intra-transaction parallelismand state encapsulation in their applications to reduce latencyand improve locality. Moreover, reactors enable a new degree offlexibility in database deployment. We present ReactDB, a systemdesign exposing reactors that allows for flexible virtualization ofdatabase architecture between the extremes of shared-nothing andshared-everything without changes to application code. Our experimentsillustrate latency control, low overhead, and asynchronicitytrade-offs with ReactDB in OLTP benchmarks

AB - The requirements for OLTP database systems are becoming evermore demanding. Domains such as finance and computer gamesincreasingly mandate that developers be able to encode complexapplication logic and control transaction latencies in in-memorydatabases. At the same time, infrastructure engineers in these domainsneed to experiment with and deploy OLTP database architecturesthat ensure application scalability and maximize resource utilizationin modern machines. In this paper, we propose a relationalactor programming model for in-memory databases as a novel,holistic approach towards fulfilling these challenging requirements.Conceptually, relational actors, or reactors for short, are applicationdefined,isolated logical actors that encapsulate relations and processfunction calls asynchronously. Reactors ease reasoning aboutcorrectness by guaranteeing serializability of application-level functioncalls. In contrast to classic transactional models, however, reactorsallow developers to take advantage of intra-transaction parallelismand state encapsulation in their applications to reduce latencyand improve locality. Moreover, reactors enable a new degree offlexibility in database deployment. We present ReactDB, a systemdesign exposing reactors that allows for flexible virtualization ofdatabase architecture between the extremes of shared-nothing andshared-everything without changes to application code. Our experimentsillustrate latency control, low overhead, and asynchronicitytrade-offs with ReactDB in OLTP benchmarks

UR - http://www.scopus.com/inward/record.url?scp=85048774436&partnerID=8YFLogxK

U2 - 10.1145/3183713.3183752

DO - 10.1145/3183713.3183752

M3 - Article in proceedings

SN - 978-450347037

SP - 259

EP - 274

BT - SIGMOD 2018 - Proceedings of the 2018 International Conference on Management of Data

A2 - Das, Gautam

A2 - Jermaine, Christopher

A2 - Eldawy, Ahmed

A2 - Bernstein, Philip

PB - Association for Computing Machinery

Y2 - 10 June 2018 through 15 June 2018

ER -