The log file cache is a pool of 8 KB buffers used to contain log file updates. You must not modify the log file cache size unless under the direction of IBM® service. Log file buffers are always written asynchronously to disk and typically need to be waited upon only when the log is becoming full, or if a file is in file synchronization (fsync).

The log file cache is stored in a data space and its default size is 16 MB. The log file cache is grown dynamically by adding one 8 KB buffer for each attached aggregate. This growth ensures that each aggregate always has one 8 KB buffer to use to record its most recent changes to file system metadata. Because log files are written asynchronously, the cache essentially allows write-behind of log files and because the cache is shared among all aggregates. Aggregates that have a higher write rate use more buffers in the cache using a least-recently-used (LRU) algorithm.

The log file cache is a write-only cache. The read hit ratio is not important. Make the log file cache large enough to avoid frequent log file I/O waits. The modify zfs,query,all command shows log file I/O waits. However, every workload is different. For example, workloads that issued fsync operations force zFS to sync the log file more frequently.