The client
program can move multiple files or directories
between the client and server before it commits the data to server
storage.
A transaction that contains
multiple files or directories is called
a transaction group. Using the TXNGROUPMAX server
option, you can specify the number of files or directories that are
contained within a transaction group. You can use the TXNGROUPMAX option
to increase performance when Tivoli® Storage
Manager writes to
tape. This performance increase can be considerable when you transfer
multiple small files.
If you increase the value of
TXNGROUPMAX by
a
large amount, monitor the effects on the recovery log. A larger value
for the
TXNGROUPMAX option can have the following
impact:
- Affect the performance of client backup, archive, restore,
and
retrieve operations.
- Increase utilization of the recovery
log, as well as increase
the length of time for a transaction to commit.
Also consider
the number of concurrent sessions to be run. It
might be possible to run with a higher
TXNGROUPMAX value
with a few clients running. However, if there are hundreds of clients
running concurrently, you might need to reduce the
TXNGROUPMAX to
help manage the recovery log usage and support this number of concurrent
clients. If the performance effects are severe, they might affect
server operations. See
Monitoring the database and recovery log for
more information.
The following examples show how the
TXNGROUPMAX option
can affect performance throughput for operations to tape and the recovery
log.
- The TXNGROUPMAX option is set to 20.
The MAXSESSIONS option,
which specifies the maximum number of concurrent client/server sessions,
is set to 5. Five concurrent sessions are processing, and each file
in the transaction requires 10 logged database operations. This would
be a concurrent load of:
20*10*5=1000
This represents
1000 log records in the recovery log. Each time a transaction commits
the data, the server can free 200 log records.
- The TXNGROUPMAX option
is set to 2000. The MAXSESSIONS option
is set to 5. Five concurrent sessions are processing, and each file
in the transaction requires 10 logged database operations, resulting
in a concurrent load of:
2000*10*5=100 000
This
represents 100 000 log records in the recovery log. Each time a transaction
commits the data, the server can free 20 000 log records.
Remember: Over time and as transactions
end, the recovery log can release the space that is used by the oldest
transactions. These transactions complete, and the log space usage
increases.
Based on the previous two examples, five concurrent
transactions
with a TXNGROUPMAX setting of 2000 consume much more
space in the recovery log. This increase in log space usage also increases
the risk of running out of recovery log space.
The following
table shows a comparison of the examples of the preceding
TXNGROUPMAX settings.
This example becomes more significant if a given log record takes
100 bytes.
Table 1. Example of log bytes that are consumed by
five concurrent sessions| TXNGROUPMAX Setting |
Number of Log Bytes Consumed |
|---|
| TXNGROUPMAX=20 |
100,000 |
| TXNGROUPMAX=2000 |
10,000,000 |
You can use several server options to tune server performance and
reduce the risk of running out of recovery log space:
- Use
the THROUGHPUTTIMETHRESHOLD and THROUGHPUTDATATHRESHOLD options
with the TXNGROUPMAX option to prevent a slower performing
node from holding a transaction open for extended periods.
- Increase
the size of the recovery log when you increase the TXNGROUPMAX setting.
Evaluate the performance and characteristics of each node
before
increasing the TXNGROUPMAX setting. Nodes that have
only a few larger objects to transfer do not benefit as much as nodes
that have multiple, smaller objects to transfer. For example, a file
server benefits more from a higher TXNGROUPMAX setting
than does a database server that has one or two large objects. Other
node operations can consume the recovery log at a faster rate. Be
careful when increasing the TXNGROUPMAX settings
for nodes that often perform high log-usage operations. The raw or
physical performance of the disk drives that are holding the database
and recovery log can become an issue with an increased TXNGROUPMAX setting.
The drives must handle higher transfer rates to handle the increased
load on the recovery log and database.
You can set the TXNGROUPMAX option
as a global
server option value, or you can set it for a single node. For optimal
performance, specify a lower TXNGROUPMAX value (between
4 and 512). Select higher values for individual nodes that can benefit
from the increased transaction size.