Title: Improving Productivity with Fine-Grain Compiler-Based Checkpointing

Authors: Chuck Zhao*, J. Gregory Steffan*, Cristiana Amza*, and Allan
         Kielstra** *University of Toronto **IBM Toronto
Speaker: Chuck Zhao

Abstract:

Checkpointing support allows program execution to roll-back to an earlier
program point, discarding any modifications made since that point.  Existing
software-based checkpointing methods are mainly libraries that snapshot all of
working-memory, and hence have prohibitive overhead for many potential
applications. Hardware-based checkpointing support could be more selective, but
is not currently available in commodity systems.

In this talk we present a light-weight, fine-grain checkpointing framework
implemented entirely in software through compiler transformations and
optimizations. A programmer can specify arbitrary checkpoint regions via a
simple API, and the compiler automatically transforms the code to implement the
checkpoint at the granularity of individual stores. Optimization passes reduce
the checkpoint size and hence improve performance by identifying and
eliminating redundant backup operations---in particular by analyzing memory
allocation sites, arrays, and system calls, and then performing
checkpoint-specific inlining, redundancy elimination, and hoisting passes.

To demonstrate the utility of compiler-based checkpointing we explore two
application areas. First, we investigate its application to debugging, in
particular by providing the ability to rewind to an arbitrarily-placed point in
a buggy program's execution---eliminating the need to restart a program from
the beginning for every debug attempt.  A study using BugBench applications
shows that our compiler-based approach is more than 100X less overhead than
full-process checkpointing.  Second, we demonstrate that compiler-based
checkpointing support can be leveraged to free the programmer from manually
implementing and maintaining software rollback mechanisms when coding a
backtracking algorithm, with runtime overhead of only 15% compared to the manual
implementation.