Document Details

Document Type : Thesis 
Document Title :
INTEGRATING STATIC AND DYNAMIC ANALYSIS TECHNIQUES FOR DETECTING DYNAMIC ERRORS IN MPI PROGRAMS
عمل تكامل بين طرق التحليل الثابت و التحليل الديناميكي للكشف عن الأخطاء الديناميكية في برامج MPI
 
Subject : Faculty of Computing and Information Technology 
Document Language : Arabic 
Abstract : Message passing interface (MPI) is the de-facto standard for programming high performance computing applications and it is ready for scaling to extreme scale system with millions of nodes and billions of cores with this huge number of components MPI will be error prone. Many types of errors can occur with MPI implementation such as deadlock and message condition. Testing and model checking have important value which is to find errors in programs. Further, if no errors are to be found these techniques will increase the confidence that the program is correct. Testing tools can assist application developers in the detection of such errors. This thesis contributes to the early detection of some scenarios of errors during static analysis by defining a novel representation of the target application based on stack structures. The representation is an extension of the one used by the clang compiler. This fine-grained representation allows for analyzing the flow of concurrent messages being exchanged, which is important for deadlock errors and race conditions detection. We detect these kinds of errors in point-to-point and collective communication. In a broader context, this thesis aims to improve the performance of error detection in MPI applications by integrating static analysis and dynamic analysis, where potential problematic constructs that are reported by the static part of our tool are further checked during program execution. Hence, we focus our analysis to consider only the highlighted paths, to be able to reduce time overhead of dynamic analysis. Several mini-programs are selected from a benchmark that contain examples of the errors identified by our work. These mini-programs are then tested by our tool. The experimental results show that our tool is capable of finding deadlocks and race conditions. 
Supervisor : Dr. Mai Ahmed Fadel 
Thesis Type : Master Thesis 
Publishing Year : 1439 AH
2018 AD 
Added Date : Monday, July 2, 2018 

Researchers

Researcher Name (Arabic)Researcher Name (English)Researcher TypeDr GradeEmail
رنا عبد الرزاق النمرAlnemari, Rana AbdulRazaqResearcherMaster 

Files

File NameTypeDescription
 43551.pdf pdf 

Back To Researches Page