Synchronization of independent clocks

Abstract

A system and method to synchronize independent local clocks in multi-core processing system are disclosed. A shared counter or a shared memory / file is provided to establish a partial happened-before relationship (e1<e2 in the happened-before order if we know that the event e1 happened before the event e2) and a synchronizer device is utilized to generate a global time of events in threads or processes. The synchronizer device estimates each clock's start time and approximates elapsed time between events in threads or processes by executing an all-pair shortest-path algorithm

Description

CROSS-REFERENCE TO RELATED APPLICATIONS[0001]This application is a U.S. continuation application which is filed under 35 U.S.C. 111(a) and claims the benefit under 35 U.S.C. 120 of an allowed U.S. patent application (U.S. patent application Ser. No. 11 / 958,918), filed on Dec. 18, 2007, the entire contents of which are hereby incorporated by reference.BACKGROUND OF INVENTION[0002]1. Field or invention[0003]The present invention relates to a clock synchronization system and method in a multi-core processing system.[0004]2. Description of the Prior Art[0005]A problem arises, when tracing programs that run on a multi-core machine having a plurality of processing cores where clocks on each core are not synchronized. This absence of consistent time across the processes complicates program debugging, performance analysis, and visualization. For example, the problem arises in the following situations:[0006]At runtime, at each point, each process has access to a clock, but not all processes ...

AI Technical Summary

Benefits of technology

[0017]The proposed system and method do not require cooperation at runtime between different nodes, making it applicable to non-preemptive environment such as Cell processor's SPE (Synergistic Processing Element: a type of Cell processor's cores). It has very low runtime overhead and code space requirements. It is completely offline, so that the time synchronization overhead happens completely during post-processing. It can recover from errors due to imprecision in local time measurements and differences in the cost of accessing the synchronizing resources.

Problems solved by technology

A problem arises, when tracing programs that run on a multi-core machine having a plurality of processing cores where clocks on each core are not synchronized.

This absence of consistent time across the processes complicates program debugging, performance analysis, and visualization.

Some cores cannot run background processes.

Therefore, it is not sufficient for performance analysis and visualization.

Therefore, it cannot be used with the cores that cannot run background processes.

Images