Debugging Support for Charm

1
Debugging Support for Charm
Rashmi Jyothi
2
Support in Charm Overview

• Ability to freeze/unfreeze the program
• Set/remove breakpoints
• View contents of array elements, messages in
  queues, readonlies, registration tables
• Attach processes on specified number of processor
  elements to a sequential debugger during program
  execution
• Basic record/replay functionality

3
Debugger Approach
Uses the CCS module in Charm
The program to be debugged is the parallel
server while the debugger is the client
Charm program
Parallel Debugger
ccs_set_break_point
ccs_debug_startgdb
. .
4
Debugger Approach (contd)

• CpdList allows to retrieve the contents of the
  chare table, messages in queues etc of a running
  program .
• Ccs handler written which returns the pupped data
  for a particular CpdList, to the client
• Lists registered to retrieve items in chare
  table, entry table, CmiLocalQueue, CsdSchedQueue
  etc
• At time of initialization, pup function to be
  used for each of the CpdLists registered.
• PUPtoText PUPsizerText used to return
  displayable strings

5
Record Replay

• This mechanism is useful to find bugs that occur
  due to the order in messages are processed
• Program linked with tracemode recordreplay
• ltsrcpe env-size sequence nogt written to a file
  for each pe when record runtime option used
• replay reruns the program with the same message
  ordering sequence.
• Tested for cases not using migration.

6
Thank you