Browse Prior Art Database

Improving Performance of Visual Trace Analysis using Abstract Time-Lines

IP.com Disclosure Number: IPCOM000115103D
Original Publication Date: 1995-Mar-01
Included in the Prior Art Database: 2005-Mar-30
Document File: 4 page(s) / 467K

Publishing Venue

IBM

Related People

Berry, RF: AUTHOR [+3]

Abstract

"Tracing" a computer system is the recording of the events or states which the system executed. Trace chronologies are used to analyze and debug the performance of computer systems. Often, the chronology can be ordered visually on one or more time-lines marked with various icons, colors and shapes to represent the events contained in the trace record. There are a number of visual tools which attempt to order trace data in these ways as a means to for giving performance analysts powerful ways to study causal relationships of trace events.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 40% of the total text.

Improving Performance of Visual Trace Analysis using Abstract Time-Lines

      "Tracing" a computer system is the recording of the events or
states which the system executed.  Trace chronologies are used to
analyze and debug the performance of computer systems.  Often, the
chronology can be ordered visually on one or more time-lines marked
with various icons, colors and shapes to represent the events
contained in the trace record.  There are a number of visual tools
which attempt to order trace data in these ways as a means to for
giving performance analysts powerful ways to study causal
relationships of trace events.

      These traces are especially powerful when the traces contain,
as they often do, the equivalent of hundreds or even
tens-of-thousands of pages of textual information.  In fact, without
such tools, only small fractions of such traces are ever studied with
any success.

      PieScope is a visual performance analysis tool which builds
various time-lines of a trace.  Typically, each time-line corresponds
to a thread-of-execution such a process or thread.  PieScope shows
many of the events belonging to a process or thread, but for the
purposes of this disclosure, only the scheduling and locking behavior
of the process or thread are important.

      PieScope shows the lock acquisition patterns of processes and
threads using colored bars laid along each time-line.  Where ever a
colored locking bar lies on a time-line the corresponding process or
thread held a lock.  This kind of representation has proved valuable
in understanding and analyzing locking behavior.  PieScope shows how
these processes or threads are scheduled using alternating patterns
of colored and uncolored bars along each time-line.  Where ever a
colored bar lies on the time-line, the corresponding process or
thread was running on a cpu.  This kind of representation has proved
valuable in understanding and analyzing computer scheduling
performance.

      When analyzing the scheduling or locking performance of system,
performance analysts inevitably want to know queue lengths and
waiting times (service times in queuing theory terminology).  Often,
all they are able to gleam from a trace are statistical
representations of queue lengths and wait times.  However, analysts
occasionally need to know queue lengths and wait times precisely at
arbitrarily selected parts of a trace.  If, for example, an analyst
discovers a period in the trace where there are alternating patterns
of processes or threads running for very short periods of times
followed by one or two processes or threads running for very long
period of times, the analyst will want to know the precise behavior
the queues during the time leading up to and during the unusual
scheduling behavior.

      The problem is that although information about schedule and
lock queue lengths and wait times is available in the trace, it is
tedious to extract when viewed on per process...