Scheduling system for distributed transaction processing

Abstract

This chapter describes a method of integration of IT systems working with many shared database storage in distributed environment. Following sections describe new elements in distributed environment: agents, naming service and coordination service. Every IT system pass all transactions on individual agent, that suspend their execution. Agent register suspended transaction in naming service, obtaining binary representation of their resource data. Registered transaction is requested by agent in central coordination service. Detection of conflict between transactions is realized by coordinator service and assures data integrity by controlling the execution of requested transactions. This feature eliminates deadlock and possibility of transactions rollbacks due to data conflicts. Besides theoretical description authors introduce formula for transaction conflict detection. The algorithm of conflicts detection is designed to be executed on a special hardware accelerator based on a systolic array.