edu.ufl.hcs.ppw.pdm.highlevel.TraceMerger Class Reference

This class sits on top of the trace interval reader, providing a single stream of IntervalRecords (sorted on end time) instead of N streams, one per thread. More...

List of all members.

Classes
class	ThreadRecords
	Class to represent each thread's bucket of records. More...
Public Member Functions
	TraceMerger (HighLevelPDM pdm) throws IOException
	Create a new trace record merger for the given PDM.
IntervalRecord	getNextInterval () throws IOException
	Returns the next record across all threads.
Static Public Member Functions
static void	main (String[] args) throws IOException
	test harness
Package Attributes
int	threads
HighLevelPDM	pdm
ThreadRecords[]	threadRecs
Static Package Attributes
static final int	CHUNKS = 64
	Number of records to read in at once, per thread.

---

Detailed Description

This class sits on top of the trace interval reader, providing a single stream of IntervalRecords (sorted on end time) instead of N streams, one per thread.

It is basically an on-line merge.

As with the TraceIntervalReader, do not use the underlying raw trace record accesses of the HighLevelPDM unless you want this to break.

Author:

Adam Leko

---

Member Function Documentation

IntervalRecord edu.ufl.hcs.ppw.pdm.highlevel.TraceMerger.getNextInterval

(

)

throws IOException

Returns the next record across all threads.

If no more are available, this returns null.

TODO: This could be improved a lot by using a min heap instead of a dumb linear search.

References edu.ufl.hcs.ppw.pdm.highlevel.IntervalRecord.getEndRecord(), edu.ufl.hcs.ppw.pdm.TraceRecord.getTimestamp(), edu.ufl.hcs.ppw.pdm.highlevel.TraceMerger.TraceMerger.ThreadRecords.peekNextInterval(), and edu.ufl.hcs.ppw.pdm.highlevel.TraceMerger.TraceMerger.ThreadRecords.popNextInterval().

Referenced by edu.ufl.hcs.ppw.exports.SLOG2TraceConverter.convert(), and edu.ufl.hcs.ppw.pdm.highlevel.TraceMerger.main().

---

Member Data Documentation

final int edu.ufl.hcs.ppw.pdm.highlevel.TraceMerger.CHUNKS = 64 [static, package]

Number of records to read in at once, per thread.

This is an important parameter; larger values reduce costly disk seeks, but take up more memory (especially with a large number of threads).

Referenced by edu.ufl.hcs.ppw.pdm.highlevel.TraceMerger.TraceMerger.ThreadRecords.fillIfNecessary().

---

The documentation for this class was generated from the following file:

• javasrc/src/edu/ufl/hcs/ppw/pdm/highlevel/TraceMerger.java