Backtracking for Distributed Model Execution

1
Backtracking forDistributed Model Execution

• Motivation
• Distributed model execution requires
  backtracking. When a component receives a message
  tagged with a time in its past, it must rollback
  and process the message in the correct sequential
  order.
• Incremental backtracking aims to capture every
  change in model states. Every time an actor
  assigns a new value to its private non-static
  field (state), the old value must be remembered
  so that a rollback can be issued later.
• Automatic Java Source Refactoring
• It is boring for the programmers to manually
  capture all the changes, so we take a refactoring
  approach.
• An Eclipse plug-in refactors actor source code
  as it is being written. A command-line tool is
  implemented to refactor a list of sources at a
  time.
• Backtracking with Checkpoint Objects
• A checkpoint object monitors a group of objects
  to be checkpointed and rolled back at the same
  time.
• A timestamp is the handle of a checkpoint.
• The rollback method of a checkpoint object rolls
  the state back to a previous timestamp.
• Modified SDF Director
• In its original design, SDF does not conform to
  the prefire-fire-postfire semantics.
• The modified SDF director conforms to the
  prefire-fire-postfire semantics by recording a
  checkpoint before the first fire.
• SDF composite actors can now be embedded in CT
  and provide the correct semantics in a clean way.

package mypackage import java.io.InputStream /
A buffer of predefined size for an input
stream. / public class StreamBuffer /
Construct a buffer with an input stream.
/ public StreamBuffer(InputStream
stream) _stream stream
/ Return the next byte in the buffer.
If the buffer is empty, more bytes are
read from the input stream. / public
byte nextByte() throws Exception if
(_pointer gt SIZE)
_stream.read(_buffer) _pointer 0
return _buffer_pointer
/ Size of the buffer.
/ public static final int SIZE 20
// The buffer. private byte _buffer new
byteSIZE // The current read position
in the buffer. private int _pointer SIZE
// The input stream to be read.
private InputStream _stream
package refactored.mypackage import
java.io.InputStream import java.lang.Object impo
rt ptolemy.backtrack.Checkpoint import
ptolemy.backtrack.Rollbackable import
ptolemy.backtrack.util.CheckpointRecord import
ptolemy.backtrack.util.FieldRecord / A
buffer of predefined size for an input stream.
/ public class StreamBuffer implements
Rollbackable protected Checkpoint
CHECKPOINT new Checkpoint(this) /
Size of the buffer. / public
static final int SIZE 20 // The buffer.
private byte _buffer new byteSIZE
// The current read position in the buffer.
private int _pointer SIZE // The input
stream to be read. private InputStream
_stream / Construct a buffer
with an input stream. / public
StreamBuffer(InputStream stream)
ASSIGN_stream(stream) /
Return the next byte in the buffer. If the
buffer is empty, more bytes are read from
the input stream. / public byte
nextByte() throws Exception if
(_pointer gt SIZE)
_stream.read(BACKUP_buffer())
ASSIGN_pointer(0) return
_bufferASSIGNSPECIAL_pointer(11, _pointer)
// Extra fields and methods omitted.
...