This screen allows you to display I/O statistics for each file in
the database. When you select "IO Statistics (by file)" from the
display menu, Oracle Rdb displays a list of files that comprise
the database and for which you can choose to view statistics.
With the exception of the all data/snap files screen, each screen
shows the I/O activity for a specific database file. The all data
/snap files screen shows a summary of I/O activity for all data
and snapshot files.
The information in this screen applies from the time that your
Performance Monitor session began, or since the accumulators were
last reset (using the [R]eset option). The following information
is displayed:
o total I/Os
The total number of I/O operations to the file being
displayed, broken down to the number of synchronous read,
synchronous write, extend, asynchronous read, and asynchronous
write operations. This number does not include creation or
truncation operations.
o rate per second
This section provides information on the rate at which
I/O operations are performed by the database. A single I/O
operation may access several blocks of information; thus, the
number of I/O operations does not correlate directly to the
volume of data manipulated. However, if the file is on a disk
by itself and this number is high (greater than 30 for RA8n
disks), the file may be a good candidate for partitioning.
This number also indicates how much use the file is getting,
and if the file should be on its own disk.
The information is partitioned into the following categories:
o max.
This category indicates the maximum number of operations
per second attained for synchronous read, synchronous
write, extend, asynchronous read, and asynchronous
write operations. The max. category is typically used to
differentiate peaks in performance from the system-wide
average. Note that no total figure is displayed for this
category.
o cur.
This category indicates the current number of operations
per second attained for synchronous read, synchronous
write, extend, asynchronous read, and asynchronous write
operations. This information is computed using the number
of I/O operations performed during the screen-update time
interval.
o avg...
This category indicates the average number of operations
per second attained for synchronous read, synchronous
write, extend, asynchronous read, and asynchronous
write operations. The avg. category is typically used
to determine overall system activity. Note that no total
figure is displayed for this category.
o total count
This category indicates the total number of synchronous
read, synchronous write, extend, asynchronous read, and
asynchronous write operations performed by the database.
The total count category serves as the basis for the Average
Per I/O categories. It is a good indicator of how much I/O
activity there is for a file and for comparing one file's
activity with other files in the database.
o average per trans
This category indicates the total number of synchronous read,
synchronous write, extend, asynchronous read, and asynchronous
write operations, based on the total number of application
transactions committed by all processes attached to the
database. The total number of committed transactions is
not displayed. The processing within a transaction and the
duration of a transaction for this category are determined
entirely by the application.
The average per trans category can indicate whether or not
certain transactions incur a high number of I/O operations to
accomplish a task. If you notice a number that is higher than
you expected, you should investigate further by using other
Performance Monitor options.
o blocks transferred
This section provides information on the volume of data
manipulated (transferred) by the database. Depending on the
operation, Oracle Rdb attempts to make an I/O operation do
as much work as possible by transferring as many blocks as
possible. The blocks transferred statistic shows how much work
is being done for each I/O operation. A number less than the
buffer size indicates that either SPAM pages are used heavily
(because they fill only one page), or there are contention
problems because Oracle Rdb resolves contention problems by
writing and reading pages rather than full buffers.
The information is partitioned into the following categories:
o avg. per I/O
This category indicates the average number of blocks
transferred for synchronous read, synchronous write,
extend, asynchronous read, and asynchronous write
operations, based on the total number of I/O operations
that have been performed by the database.
o total...
This category indicates the total number of blocks
transferred for synchronous read, synchronous write,
extend, asynchronous read, and asynchronous write
operations.
o stall time (x100)
This section provides information on how long processes wait
for the I/O operations to complete. Processes typically stall
due to contention for the disk by other processes. Stall time
is measured in hundredths of a second but is displayed as
whole numbers. Thus, a displayed value of 100 represents 1
second of stall time. A large number can indicate excessive
I/O activity to a file and that requests are becoming stacked
in a queue. This indicates the file is a good candidate for
partitioning or being placed on another disk. Normally, RA8n
disks should complete an I/O operation in 2/100 to 4/100 of a
second, while RD5n disks should complete an I/O operation in
5/100 to 8/100 of a second.
The information is partitioned into the following categories:
o avg. per I/O
This category indicates the average time stalled for
synchronous read, synchronous write, extend, asynchronous
read, and asynchronous write operations based on the number
of I/O operations that have been performed.
When you are displaying statistics for all data/snap files,
the value for the average stall time per I/O field may not
be what you expect if you have displayed the average stall
time per I/O for the individual .rda and .snp files in the
database. For example, it is possible for the average stall
time per I/O for the all data/snap files screen to be 1.1
when the average stall times per I/O for the .rda and .snp
files are 8.3, 5.7, 6.4, and 6.8, respectively.
Oracle Rdb issues write I/O operations in parallel,
asynchronously (this is a batch-write mechanism). This
means that the average stall time per I/O for the all data
/snap files screen is not the "average of the averages"
shown for the individual .rda and .snp files; this would
imply that the I/Os completed serially. Rather, the total
I/O stall time is "the average of the averages divided
by the number of I/Os;" the average for the all data/snap
files screen is usually a fraction of the individual file
averages because the stall time is amortized across all
I/Os issued in parallel.
For example, assume that Oracle Rdb does a batch-write
operation (all in parallel, of course) to four storage
areas. Assume that each individual storage area's I/O
operation takes 20 milliseconds. If the I/Os were done
serially, then the average stall time for all storage areas
would be 20 milliseconds. However, because the I/Os are
done in parallel, the average for all areas is actually 5
milliseconds (20 milliseconds divided by 4 I/O operations).
o total...
This category indicates the total time stalled for
synchronous read, synchronous write, extend, asynchronous
read, and asynchronous write operations.
Note that the accumulators on this screen can be reset using the
[R]eset option.