This section describes fsflush and related tunables.
On every invocation, fsflush does the following:
Flushes dirty file system pages over a certain age to disk.
Examines a portion of memory and causes modified pages to be written to their backing store. Pages are written if they are modified and do not meet one of the following conditions:
Associated with a swap device
Currently involved in an I/O operation
The net effect is to flush pages from files which are mmap(ed) with write permission and which have actually been changed.
Pages are flushed to backing store but left attached to the process using them. This will simplify page reclamation when the system runs low on memory by avoiding delay for writing the page to backing store before claiming it, if the page has not been modified since the flush.
Writes file system metadata to disk. This write is done every nth invocation, where n is computed from various configuration variables. See tune_t_fsflushr and Where to Find Tunable Parameter Information for details.
Frequency of invocation, whether the memory scanning is executed, whether the file system data flushing occurs, and the frequency with which it will occur are configurable.
For most systems, memory scanning and file system metadata syncing are the dominant activities for fsflush. Depending on system usage, memory scanning can be of little use or consume too much CPU time.
1 to MAXINT
If the value is less than or equal to zero, the value is reset to 5 and a warning message is displayed. This check is only done at boot time.
See autoup below.
Along with tune_t_flushr, autoup controls the amount of memory examined for dirty pages in each invocation and frequency of file system sync operations.
The value of autoup is also used to control whether a buffer is written out from the free list. Buffers marked with the B_DELWRI flag (file content pages that have changed) are written out whenever the buffer has been on the list for longer than autoup seconds. Increasing the value of autoup keeps the buffers around for a longer time in memory.
1 to MAXINT
If autoup is less than or equal to zero, it is reset to 30 and a warning message is displayed. This check is only done at boot time.
autoup should be an integer multiple of tune_t_fsflushr. At a minimum, autoup should be at least 6 times tune_t_fsflushr. If not, excessive amounts of memory will be scanned each time fsflush is invoked.
(total system pages x tune_t_fsflushr) should be greater than or equal to autoup to cause memory to be checked if dopageflush is non-zero.
There are several potential situations for changing autoup and or tune_t_fsflushr:
Systems with large amounts of memory—In this case, increasing autoup reduces the amount of memory scanned in each invocation of fsflush.
Systems with minimal memory demand—Increasing both autoup and tune_t_fsflushr reduces the number of scans made. autoup should be increased also to maintain the current ratio of autoup / tune_t_fsflushr.
Systems with large numbers of transient files (for example, mail servers or software build machines)—If large numbers of files are created and then deleted, fsflush might unnecessarily write data pages for those files to disk.
Controls whether memory is examined for modified pages during fsflush invocations. In each invocation of fsflush, the number of memory pages in the system is determined (it might have changed because of a dynamic reconfiguration operation). Each invocation scans (total number of pages x tune_t_fsflushr) / autoup pages.
0 (disabled) or 1 (enabled)
If the system page scanner rarely runs, indicated by a value of 0 in the sr column of vmstat output.
Controls whether file system metadata syncs will be executed during fsflush invocations. Syncs are done every Nth invocation of fsflush where N= (autoup / tune_t_fsflushr). Because this is an integer division, if tune_t_fsflushr is greater than autoup, a sync will be done on every invocation of fsflush because the code checks to see if its iteration counter is greater than or equal to N. Note that N is computed once on invocation of fsflush. Later changes to tune_t_fsflushr or autoup will have no effect on the frequency of sync operations.
0 (disabled) or 1 (enabled)
When files are frequently modified over a period of time and the load caused by the flushing perturbs system behavior. Files whose existence, and therefore consistency of state does not matter if the system reboots, are better kept in a TMPFS file system (for example, /tmp). Inode traffic can be reduced on systems running the Solaris 7, 8, or 9 releases by using the mount -noatime option. This option eliminates inode updates when the file is accessed.
A system engaged in realtime processing might want to disable this option and use explicit application file syncing to achieve consistency.