Logging System Summary Fields
The Logging System Summary screen has the following fields:
Database adds
The number of databases added to the logging system.
Database removes
The number of databases removed from the logging system. This means the last user of a database has disconnected from Ingres. The number of databases currently open is the difference between Database adds and Database removes.
Log writes
The number of log writes. This is a memory-to-memory write.
UNIX: Shows the number of log writes into shared memory.
VMS: Shows the number of log writes into the non-paged pool.
Write I/Os
The number of write I/Os actually done to the log file. This is a physical write.
Read I/Os
The number of read I/Os actually done to the log file. This is a physical read.
Log forces
The number of log force requests. These occur under the following circumstances:
• The buffer manager forces a specific log page to disk. This occurs during the consistency point (for fast commit) or before a transaction commit.
• The buffer manager forces the last log buffer in the logging system to disk. This occurs normally during the close of a table.
• The RCP or DBMS Server forces the last log record of a transaction to disk before recovering the transaction.
Check commit timer
The number of times that the timer, associated with the group commit, is completed. This does not necessarily mean that a write to the log file occurs. A write does not occur if the log buffer, that initiated the timer, is full. This timer is activated only when there are multiple transactions active in the logging system that can cause delays for users because of group commit.
The timer wakes up every 10 milliseconds and checks if there is a buffer waiting to be forced to the log file. If a buffer is found, the timer waits for 20 additional milliseconds before calling the routine to write out the buffer.
Timer write
The number of times a group commit timer check resulted in a group commit.
Inconsistent DB
The number of inconsistent database occurrences. This number must be zero. If it is not, check the errlog.log file.
UNIX: The errlog.log file is in $II_SYSTEM/INGRES/files
VMS: The errlog.log file is in II_SYSTEM:[INGRES.FILES]
Tx begins
The number of transactions started.
Tx ends
The number of commits or transactions ended. This value represents user-initiated transaction ends such as commits, rollbacks, and interrupts. It does not include system-generated transaction ends such as LOG-FULL. The difference between Transaction begins and Transaction ends is the number of current transactions.
Log waits
The number of times any event wait condition requires a log buffer write to stall. These events encompass things like LOG-FULL, CP writing, RECOVERY, archiving required, FREE WAIT for log buffer, OPENDB wait, log buffer SPLIT WAIT, and wait for log I/O to complete (the log buffer being written to the log file).
Log split waits
The number of times a log split operation is delayed due to the lack of free log buffers. This must be watched because the logging system cannot proceed with the log record split until a free buffer is available. You can reduce this by increasing the number of log buffers or increasing their size.
Log free waits
The number of times all the log buffers are either in force mode or unavailable for writing. Only one log buffer is currently written to at a time. If this state is frequent (more than five per minute during busy times), an increase in the number of log buffers is the solution. Remember that an increase in the number of buffers requires more memory (number_of_buffers x buffer_size). This condition affects throughput to the log file. For more information, see the System Administrator Guide.
Log stall waits
The number of times stalled while making requests to the logging system. This occurs while consistency points are being taken or during LOG-FULL conditions. This is acceptable (not in the LOG-FULL case) and these types of stalls only last for a fraction of a second. If 100 sessions are stalled for one event, this count is incremented by 100. Only stalls due to LOG-FULL events are reflected in this value.
For more information on log stall waits see,
Logging System Header Screen. Depending upon the causes of a Log stall wait, the following statuses are shown in the Logging System Header screen:
LOGFULL, FORCE_ ABORT
The log file is full.
LOGFULL, CPFLUSH
The log file is overly full while executing a consistency point (CP). The CP must complete before log space can be released. The status LOGFULL, ARCHIVE is also be displayed.
LOGFULL, ARCHIVE
The log file becomes full and journaled transactions must be archived to free up log space. When the archiver is done, log file space is released.
Log group commit
The number of times that multiple transactions are participating in a log buffer flush to the log file. The value is incremented every time a write to disk completes a group commit (piggyback write).
Log group count
The number of transactions that are participating in the flush to the log file. If one group commit event writes on behalf of ten sessions (threads), this number is incremented by ten. The ratio between this number and the Log group commit indicates how effective group commit is (for example, group count/group commit indicates the number of log write requests by threads that are satisfied per group commit write).
Kbytes written
The number of kilobytes written to the log file.
Avg. Buffer Utilization
The average amount of data in each log buffer written to the log, expressed as a percentage of a log buffer. The calculation used is (Kbytes written * 100) / (Log write I/Os * buffer size in KB).