IOSTATSTABLE(1)
NAME
IOstatsTable - Produce tables summarizing information
about I/O operations from I/O detail trace records
SYNOPSIS
IOstatsTable [-processor number] [-fileID id] tracefile
DESCRIPTION
IOstatsTable generates a report containing two tables of
information about application I/O activity summarized by
request type from detailed I/O trace records in the input
SDDF trace file. The necessary trace event records are
produced by the I/O extension to the Pablo trace library
by default whenever the I/O extension has been initialized
and individual I/O calls have been instrumented.
As IOstatsTable is running, it periodically displays the
number of input trace packets (records) processed to stan-
dard error. The report output is directed to standard
out.
If the input file does not include any I/O detail trace
records, the report will be generated without error but
will contain all zeros. For the report to work as
intended, there must be I/O detail trace records in the
input SDDF file. If the programmer elects not to collect
I/O detail trace records for the entire application execu-
tion, the report must be interpreted with that in mind.
OPTIONS
-processor number
Only I/O activity occurring on the specified pro-
cessor is included in the report. A comment con-
taining the processor number is written to standard
error when IOstatsTable begins execution. The pro-
cessor number is also included in the report itself
after the heading "Reported Processor:".
The default behavior is to include I/O activity
from all processors.
-fileID id
Only I/O activity associated with the specified
file, as identified by id, is included in the
report. A comment containing the file ID is writ-
ten to standard error when IOstatsTable begins exe-
cution. The file ID and file name are also included
in the report itself after the headings "Reported
File ID:" and "Reported File:".
If the input trace uses the same fileID to refer to
multiple files, the information from all files
associated with the selected ID will be included in
the report. However, the name of only one of the
files will be listed. See the section below on
"BEFORE RUNNING THE PROGRAM" and also Syn-
cIOfileIDs(1) for further discussion on how to
avoid this file ID to file name mismatch problem.
The default behavior is to include I/O activity for
all files in the report.
BEFORE RUNNING THE PROGRAM
The application to be studied must be instrumented with
the I/O extension to the Pablo trace library and should
NOT call the function disableIOdetail(). The instrumented
application is run and one or more trace files are gener-
ated - the number of trace files depends on the number of
processors used to run the application.
If there are multiple trace files, they should be merged
with the MergePabloTraces(1) program to produce a single
trace file for the execution including information from
all processors. It is possible to run IOstatsTable on
unmerged trace files, but the output will only contain
information for the single processor reported in the input
trace file.
After the trace files have been merged, you should run the
program AdjustTime(1) on the merged trace file if you
linked the application with the the raw clock version of
the Pablo trace library.
If you intend to use the -fileID option to report only
accesses to individual files, the program SyncIOfileIDs(1)
should be run on the input trace file to synchronize IDs
for files that were opened more than once by the applica-
tion. The file generated by SyncIOfileIDs ending in
".syncFiles" can be used as input to IOstatsTable. The
file ending in ".syncFiles.map" can be used to identify
the file ID of the particular file whose I/O you are
interested in viewing.
If I/O detail trace records were not produced for all I/O
operations in the application, either because some calls
were not instrumented or because detail tracing was dis-
abled for part of the execution, the report will include
only those operations for which detail records were pro-
duced and should be interpreted with that in mind.
THE REPORT
The report tabulates information from detailed I/O event
trace records. Every I/O operation for the processor(s)
and file(s) in the input tracefile that is not filtered
out by a command line -fileID or -processor switch is
treated separately - there is no attempt to correlate
operations that take place concurrently on individual
nodes of a multiprocessor system. Said another way, the
processor and file ID information in the trace event
records is used for filtering purposes, but is not consid-
ered beyond that as the statistics are generated for this
report.
The input trace file is listed near the top of the report.
The reported first and last I/O operation timestamps are
based on I/O requests from every processor included in the
report. The number of nodes reported is one plus the
highest processor identifier appearing in the trace file.
The reported wall-clock time is the greatest timestamp in
the trace file, and the execution time is the wall-clock
time multiplied by the number of nodes.
The first table in the report shows the following informa-
tion for each I/O request type and a summary line for all
I/O requests:
o Count: count of occurances
o %Count: percentage of occurances relative to all
reported I/O requests
o IO time: total time spent (in seconds)
o %IOtim: percentage of time spent relative to all
reported I/O requests
o %EXtim: percentage of time spent relative to exe-
cution time
o Bytes: total bytes transferred
o %Bytes: percentage of bytes transferred relative
to bytes transferred for all reported I/O requests
The second table provides a distribution of read and write
requests based on request size.
For systems which include Asynch Read and Asynch Write
information, the durations reported are for the asyn-
chronous read or asynchronous write calls and do not
include the actual time required to transfer the requested
bytes.
SEE ALSO
AdjustTime(1), FileRegionIOstats(1), IOstats(1),
IOtotalsByPE(1), MergePabloTraces(1), LifetimeIOstats(1),
SyncIOfileIDs(1), TimeWindowIOstats(1)
Ruth A. Aydt, A User's Guide to Pablo I/O Instrumentation
Ruth A. Aydt, The Pablo Self-Defining Data Format
COPYRIGHT
Copyright 1996, The University of Illinois Board of
Trustees.
AUTHOR
Ruth A. Aydt, University of Illinois
Pablo Environment Feb 22, 1996