A condition monitoring system tracks real-world variables and alerts users when a predefined condition becomes true, e.g., when enemy planes take off, or when suspicious terrorist activities and communication are detected. However, a monitoring server can easily get overwhelmed by rapid incoming data. To prevent this, we partition the condition being monitored and distribute the workload onto several independent servers. In this paper, we study the problem of how to make a partitioned system behave "equivalently" to a one-server system. We identify and formally define three desirable properties of a partitioned system, namely, orderedness, consistency, and completeness. We propose assignment-based partitioning as a solution that can handle opaque conditions and simplifies load balancing. We also look at a few typical partitioned systems, and discuss their merits using several metrics that we define. Finally, an algorithm is presented to reduce complex system configurations to simpler ones.
[1]
Leslie Lamport,et al.
Time, clocks, and the ordering of events in a distributed system
,
1978,
CACM.
[2]
David S. Rosenblum,et al.
Achieving scalability and expressiveness in an Internet-scale event notification service
,
2000,
PODC '00.
[3]
Jennifer Widom,et al.
Models and issues in data stream systems
,
2002,
PODS.
[4]
Hector Garcia-Molina,et al.
Replicated condition monitoring
,
2001,
PODC '01.
[5]
Yuanyuan Zhao,et al.
Exploitng event stream interpretation in publish-subscribe systems
,
2001,
PODC '01.
[6]
Tony Bourke.
Server Load Balancing
,
2001
.
[7]
Marcos K. Aguilera,et al.
Matching events in a content-based subscription system
,
1999,
PODC '99.
[8]
Hector Garcia-Molina,et al.
The SIFT information dissemination system
,
1999,
TODS.