| The research on computer systems which can process in parallel is ine of the important branch for the main stream of distributedparallel computing, and PVM is an excellent case of cluster. It is of theoretical significance andpractical value to improve the dynamic size scalability of cluster, decrease the difficulty in developing distributed parallel application software,and enhance system security capabilityaimed at rotecting the system from malicious fraudulence and attack. This dissertation takes that astarget and makes thorough research on PVM. The contributions of the dissertation are:1. By thoroughly analyzing the PVM source code, we demonstrate the kemel intemals ofPVM; the soft architecture, message passing, message driving and system control principles. Thecontent is of interest in developing distributed-parallel system and application software withmessage-driving based on multi-computer, and distributed memory environment.2. PVM task migration is the basic issue of the PVM dynamic size scalability. The dissertationsuggests a novel task migration protocol and implementation of a name-changing mechanism, theprotocol consists of the following: the task related to migration being halted temporarily,constantly changing the source/destination address of midway package to the new address for thetask migration;rebuikling correctly communication connection between tasks;after sending itsown data map to the poor destination task, the source task teminating itself; the destination taskgetting the data map, setting and running on it ,i.e. the source task being migrated to thedestination host and continuously running from the migration break point of program. Theimplementation methods are : extending the PVM message-driving mechanism with taskmigration protocol functions, and making the system being able to automatically process the system data structure related to task migration for transparent task migration. Advantages of the proposed taskmigration protocol are the following:(1)the task migration is independent, that implies several independens task migration activities can simultancously exist in the system; (2)the migration is transparent, that means any task migration does not affect other tasks of the PVM system;(3)once migration is completed , the system communication traffic does not increase for thereason that the migration has happened before;(4)the protocol is very simple.3. The dissertation proposes a novel C++ object distributed-parallel mechanism based on PVM. It'smsin points are:(1)mapping main aplication program and parallel-classes to PVM tasks, alsomapping a request object message to a request PVM task message;(2)making the PVM systemsupport for C++ object distributed-parallel throughextending its protocol and prmlib withobject distributed-parallel protocol functions;(3using preprocesser to separate the parallel-classes from user's job program, then change them to PVM tasks and dispatch them to (remote_ hostcomputers in PVM for compiling and running them over there;(4)finally, running the user application main program thru preprocessing task, and communicating with parallel-class task to complete the user's job.4. The dissertation proposes a new kind of Capability implementation mechanism for a trusted OSsecure kemel which can thus be used in cluster systems. F unctions of the proposedcapability mechanism are the following:(1)Since user(process) and program Capabilitycontrols are separately made, it is of good agility and scalabililty.(2)In terms of virtualizing the rootuser in OS kemel, it is possible to controlling system-call and the command used by root user oneby one, decomposing root user's privilege, and forbidding user to login system with root status. (3)Theproposed mechanism is compatible with the DAC mechanism of class C OS, no confict betweenthem exists. This feature can further enhance the system secunity. (4)The permission in the...
|
PDF Full Text Request