Kbase 12298: Before-image cluster size in Version 6.3.
| Autor |
Progress Software Corporation - Progress |
| Acesso |
Público |
| Publicação |
10/05/1998 |
|
Before-image cluster size in Version 6.3.
PROGRESS Release 6.3 introduces many new features and enhancements to
increase performance. This notebook entry describes one very important
aspect - the before-image subsystem, also known as the BI file.
All information stored in a PROGRESS database goes first into the
BI file, which can have multiple extents. Depending on the scoping
of the transaction(s), the information is then written to the
database itself. To have good performance during normal operation
(for example, when making changes to database), fine-tune the
Before-image subsystem to handle the system workload.
The default BI cluster size is 16 kb. This cluster size is based on
design specification in PROGRESS Version 5. Harddisks are now faster
and store more data per square inch. The amount of data transferring
between the disk controller and the physical disk has changed and
keeps changing to increase performance. To take advantage of all
these changes on the hardware front, PROGRESS has introduced new
features to optimize performance for the before-image subsystem.
The larger the BI cluster, the more data can be stored in the BI file
before a synchronization point, called CHECKPOINT. However, if
the process is terminated, the recovery time increases. Guidelines
for the BI cluster depend on memory, number of transactions per
second, and number of disks and sizes. The following table lists the
suggested BI cluster size according to database size.
Database Size Suggested bi cluster size
under 1 MB 16 kb (default)
1 - 10 MB 64 kb
10 - 100 MB 128 kb
100 - 500 MB 512 kb
over 500 MB 1024 kb
Use the following command to change BI cluster size:
proutil <database> -C truncate <BIfilename> -bi <cluster_size>
For example, to change the cluster size to 128 in a database called
db with a BI file called bi, type:
proutil db -C truncate bi -bi 128
The change of cluster size is put in the log file as message #1640.
To tune your before-image subsystem:
Abstract:
* Use systematic/scientific methods when doing analysis.
* Log all information in a note book.
* Set goals and define measurable objectives.
Options:
* Use Before-Image Writers.
* Use Before-Image Buffers.
* Use promon utility options:
- 9 Activity screen.
- 17 Before-Imaging (Undo/Redo Log) - type R&D @ selection.
* Use operating system utility to determine disk i/o waits for disk.
* Optimal: place the BI file on the fastest disk with low i/o waits.
Subject: Re: Why truncate a multi-volume bi?
Space in the BI file will be reused on a cluster by cluster
basis when the following conditions are met:
o Cluster is not needed to undo a transaction - no outstanding
uncommitted transactions with notes in the cluster.
o Cluster not needed for crash recovery - checkpoint for the cluster
has completed.
o A sufficient amount of time has elapsed to ensure that the
buffered I/O associated with a checkpoint has completed.
The BI file should never need to be truncated to cause space
to be reused.
Progress Software Technical Support Note # 12298