|M.Sc Student||Shulman Boris|
|Subject||Transactions in Active Environment|
|Department||Department of Industrial Engineering and Management||Supervisor||Dr. Opher Etzion|
In recent years, a substantial amount of work has been invested in active (reactive and proactive) systems. Active systems perform actions or signal alerts in response to the occurrence of events (following the ECA: Event - Condition - Action paradigm) that are signaled when changes in the environment occur. Simple event correlation systems, which are used in the network and system management, are intended to provide "root-cause analysis" by filtering out network messages that are symptoms and correlate network data to determine what problem occurred, and display alerts to an event console. When those systems are limited in their ability to define situations to strict patterns and to various sources, the Complex Event Processing (CEP) systems provide a more general solution for detecting situations (a possibly complex pattern over the event history).
Current implementations of the CEP systems do not have any transactional behavior. The lack of transactional behavior in CEP systems makes it difficult for those systems to be deployed in environments that require transactional behavior.
In this thesis we identify the various transactional needs of CEP systems and define a transaction model for those systems accordingly. Our transaction model extends the Nested Transactions model by relaxing the Isolation property and by allowing configuration of the necessary level of Atomicity. As part of a transactional model a concurrency control algorithm is defined. The defined algorithm combines optimistic and pessimistic locking techniques, and exploits semantic knowledge of the situation definitions, in order to increase the parallel execution of transactions in CEP systems