Strategy for estimating the size of the physical log

The size of the physical log depends on two factors: the rate at which transactions generate physical log activity and whether you set the RTO_SERVER_RESTART configuration parameter

The rate at which transactions generate physical log activity can affect checkpoint performance. During checkpoint processing, if the physical log starts getting too full as transactions continue to generate physical log data, the database server blocks transactions to allow the checkpoint to complete and to avoid a physical log overflow.

To avoid transaction blocking, the database server must have enough physical log space to contain all of the transaction activity that occurs during checkpoint processing. Checkpoints are triggered whenever the physical log becomes 75 percent full. When the physical log becomes 75 percent full, checkpoint processing must complete before the remaining 25 percent of the physical log is used. Transaction blocking occurs as soon as the system detects a potential for a physical log overflow, because every active transaction might generate physical log activity.

For example, suppose you have a one gigabyte physical log and 1000 active transactions. 1000 active transactions have the potential to generate approximately 80 megabytes of physical log activity if every transaction is in a critical section simultaneously. When 750 megabytes of the physical log fills, the database server triggers a checkpoint. If the checkpoint has not completed by the time the 920 megabytes of the physical log are used, transaction blocking occurs until the checkpoint completes. If transaction blocking takes place, the server automatically triggers more frequent checkpoints to avoid transaction blocking. You can disable the generation of automatic checkpoints.

The server might also trigger checkpoints if many dirty partitions exist, even if the physical log is not 75 percent full, because flushing the modified partition data to disk requires physical log space. When the server checks if the Physical Log is 75 percent full, the server also checks if the following condition is true:
(Physical Log Pages Used + Number of Dirty Partitions) >= 
(Physical Log Size * 9) /10)

For more information about checkpoint processing and automatic checkpoints, see Checkpoints.

The second factor to consider when estimating the size of the physical log depends on your use of the RTO_SERVER_RESTART configuration parameter to specify a target amount of time for fast recovery. If you are not required to consider fast recovery time, you are not requires to enable the RTO_SERVER_RESTART configuration parameter. If you specify a value for the RTO_SERVER_RESTART configuration parameter, transaction activity generates additional physical log activity.

Typically, this additional physical log activity has little or no affect on transaction performance. The extra logging is used to assist the buffer pool during fast recovery, so that log replay performs optimally. If the physical log is considerably larger than the combined sizes of all buffer pools, page flushing and page faulting occur during fast recovery. The page flushing and page faulting substantially reduce fast recovery performance, and the database server cannot maintain the RTO_SERVER_RESTART policy.

For systems with less that four gigabytes of buffer pool space, the physical log can be sized at 110 percent of the combined size of all the buffer pools. For larger buffer pools, start with four gigabytes of physical log space and then monitor checkpoint activity. If checkpoints occur too frequently and seem to affect performance, increase the physical log size.

A rare condition, called a physical-log overflow, can occur when the database server is configured with a small physical log and has many users. Following the previously described size guidelines helps avoid physical-log overflow. The database server generates performance warnings to the message log whenever it detects suboptimal configurations.

You can use the onstat -g ckp command to display configuration recommendations if a suboptimal configuration is detected.

Copyright© 2018 HCL Technologies Limited