for real world applications - iti.illinois.edu › sites › default › files › docs › ... ·...
TRANSCRIPT
Transparent Application Checkpointing and Recovery for Real World Applications
Abhishek Garg, Indian Institute of Technology, Delhi ADVISORS: Prof. R. K. Iyer, Prof. Z. T. Kalbarczyk, Nithin Nakka
I T I S U M M E R U N D E R G R A D U A T E I N T E R N P R O G R A M, 2 0 0 8
University of Illinois at UrbanaChampaign
www.iti.uiuc.edu Information Trust Institute
Goals
Background Research Plan
Ø Fault tolerance techniques enable a system to continue operation in the event of a failure or a fault.
Ø Checkpointing consists of storing the snapshot of the current application state and using it for restarting the application in the event of a failure.
Ø Rollback Recovery: the process state is reloaded to memory from the last checkpoint and the execution resumes.
Ø Reliability Micro Kernel (RMK), a loadable kernel module for providing application‐specific reliability mechanisms, is used to deploy the application checkpointing.
Ø Developing a loadable kernel module for LINUX systems:
q Checkpoints the application periodically.
q Recovers the application on a failure from the last saved checkpoint.
Ø Demonstrate the Kernel Module on some real world applications.
Ø Learning the Kernel Module Programming and writing small test modules to get familiar with it.
Ø Understanding the Reliability Micro Kernel framework (RMK) and the code implementing Checkpointing in the framework.
Ø Leveraging the RMK interface and developing the Checkpointing kernel module.
Algorithm
Research Results
Ø What is the essential memory and processor state needed to recover the application from failures?
Ø How to detect an application process failure in the kernel mode?
Ø How to make the checkpoint to be transparent to the application?
Ø Successfully implemented the Checkpointing Module in LINUX.
Ø Demonstrated Kernel Module with the real world application, GCC compiler, by:
q Running the compilation process in the background.
q Issuing a kill signal to terminate the process.
q Initiating the recovery.
Ø Demonstrated the recovery of the application with the same Process ID.
Related Work/Interaction with Other Projects
Ø Reliability Micro Kernel: An OS level framework for providing Application‐aware reliability.
Ø M‐Checkpointing: Transparent Local and In‐ place checkpointing for propagation‐limited errors.
Applications
OS
Hardware
RMK
Pin Interface
Modules
Fundamental Questions/Challenges
Checkpointing module searches for the application to be checkpointed in the
tasks list.
Process’s memory pages including general purpose registers are saved.
The process starts execution from the restored state.
The saved memory pages are reloaded in the process’s memory data space.
Error signal is caught and the recovery process is initiated.