Improved InnoDB I/O Scalability¶

Version Specific Information¶

• 5.1.53-12.4
  • Added variable innodb_log_block_size, method keep_average to innodb_adaptive_checkpoint.
• 5.1.54-12.5
  • Added value ALL_O_DIRECT to innodb_flush_method.

System Variables¶

variable innodb_adaptive_checkpoint¶

Version Info:	5.1.54-12.5 – Added
Command Line:	Yes
Config File:	Yes
Scope:	Global
Dynamic:	Yes
Variable Type:	String
Default Value:	none (1.0.5), estimate (1.0.6)
Values:	none, reflex, estimate, keep_average or 0/1/2/3 (for compatibility)

This variable controls the way adaptive checkpointing is performed. InnoDB constantly flushes dirty blocks from the buffer pool. Normally, the checkpoint is done passively at the current oldest page modification (this is called “fuzzy checkpointing”). When the checkpoint age nears the maximum checkpoint age (determined by the total length of all transaction log files), InnoDB tries to keep the checkpoint age away from the maximum by flushing many dirty blocks. But, if there are many updates per second and many blocks have almost the same modification age, the huge number of flushes can cause stalls.

Adaptive checkpointing forces a constant flushing activity at a rate of approximately [modified age / maximum checkpoint age]. This can avoid or soften the impact of stalls casued by aggressive flushing.

The following values are allowed:

• reflex: This behavior is similar to innodb_max_dirty_pages_pct flushing. The difference is that this method starts flushing blocks constantly and contiguously based on the oldest modified age. If the age exceeds 1/2 of the maximum age capacity, InnoDB starts weak contiguous flushing. If the age exceeds 3/4, InnoDB starts strong flushing. The strength can be adjusted by the MySQL variable innodb_io_capacity. In other words, we must tune innodb_io_capacity for the reflex method to work the best.
• estimate: If the oldest modified age exceeds 1/4 of the maximum age capacity, InnoDB starts flushing blocks every second. The number of blocks flushed is determined by [number of modified blocks], [LSN progress speed] and [average age of all modified blocks]. So, this behavior is independent of the innodb_io_capacity variable.
• keep_average: This method attempts to keep the I/O rate constant by using a much shorter loop cycle (0.1 second) than that of the other methods (1.0 second). It is designed for use with SSD cards.

In some cases innodb_adaptive_checkpoint needs larger transaction log files (innodb_adaptive_checkpoint makes the limit of modified age lower). So, doubling the length of the transaction log files may be safe.

variable innodb_adaptive_flushing¶

Command Line:	No
Variable Type:	BOOL
Default Value:	TRUE
Range:	TRUE/FALSE

This is an existing InnoDB variable used to attempt flushing dirty pages in a way that avoids I/O bursts at checkpoints. In XtraDB, the default value of the variable is changed from that in InnoDB.

variable innodb_checkpoint_age_target¶

Command Line:	Yes
Config File:	Yes
Scope:	Global
Dynamic:	Yes
Variable Type:	Numeric
Default Value:	0
Range:	0+

This variable controls the maximum value of the checkpoint age if its value is different from 0. If the value is equal to 0, it has no effect.

It is not needed to shrink innodb_log_file_size to tune recovery time.

variable innodb_enable_unsafe_group_commit¶

Version:	This variable is not needed after XtraDB 1.0.5.
Command Line:	Yes
Config File:	Yes
Scope:	Global
Dynamic:	Yes
Variable Type:	Numeric
Default Value:	0
Range:	0 - 1

This variable allows you to change the default behavior of InnoDB concerning the synchronization between the transaction logs and the binary logs at commit time. The following values are available:

• 0 (default): InnoDB keeps transactions in the same order between the transaction logs and the binary logs. This is the safer value but also the slower.
• 1: Transactions can be group-committed but the order between transactions will not be guaranteed to be kept anymore. Thus there is a slight risk of desynchronization between transaction logs and binary logs. However for servers that perform write-intensive workloads (and have RAID without BBU), you may expect a significant improvement in performance.

variable innodb_flush_log_at_trx_commit_session¶

Version Info:	5.1.49-rel11.3 – Added
Command Line:	Yes
Config File:	Yes
Scope:	Global
Dynamic:	Yes
Variable Type:	Numeric
Default Value:	3
Range:	0-3

This variable implements a session-level version of the existing global variable innodb_flush_log_at_trx_commit. It allows a session to override the global setting when a different commit mode is required by the session.

The following values are available:

• 0 / 1 / 2: These values have the same meaning as for the global innodb_flush_log_at_trx_commit
• 3 (default): The session will ignore innodb_flush_log_at_trx_commit_session and stick to the global variable

variable innodb_flush_method¶

Version Info:	5.1.54-12.5 – ALL_O_DIRECT option added
Command Line:	Yes
Config File:	Yes
Scope:	Global
Dynamic:	No
Variable Type:	Enumeration
Default Value:	fdatasync
Values:	fdatasync, O_DSYNC, O_DIRECT, ALL_O_DIRECT

This is an existing MySQL 5.1 system variable. It determines the method InnoDB uses to flush its data and log files. (See innodb_flush_method in the MySQL 5.1 Reference Manual).

The following values are allowed:

• fdatasync: Use fsync() to flush both the data and log files.
• O_SYNC: Use O_SYNC to open and flush the log files; use fsync() to flush the data files.
• O_DIRECT: Use O_DIRECT to open the data files and fsync() system call to flush both the data and log files.
• ALL_O_DIRECT: use O_DIRECT to open both data and log files, and use fsync() to flush the data files but not the log files. This option is recommended when InnoDB log files are big (more than 8GB), otherwise there might be even a performance degradation. Note: When using this option on ext4 filesystem variable innodb_log_block_size should be set to 4096 (default log-block-size in ext4) in order to avoid the unaligned AIO/DIO warnings.

variable innodb_flush_neighbor_pages¶

Command Line:	Yes
Config File:	Yes
Scope:	Global
Dynamic:	Yes
Variable Type:	Numeric
Default Value:	1
Range:	0-1

This variable specifies whether, when the dirty pages are flushed to the data file, the neighbor pages in the data file are also flushed at the same time or not. The following values are available:

• 0: Disables the feature
• 1 (default): Enables the feature

If you use a storage which has no “head seek delay” (e.g. SSD or enough memory for write buffering), 0 may show better performance.

variable innodb_log_block_size¶

Command Line:	Yes
Config File:	Yes
Scope:	Global
Dynamic:	No
Variable Type:	Numeric
Default Value:	512
Units:	Bytes

This variable changes the size of transaction log records. The default size of 512 bytes is good in most situations. However, setting it to 4096 may be a good optimization with SSD cards. While settings other than 512 and 4096 are possible, as a practical matter these are really the only two that it makes sense to use. Clean restart and removal of the old logs is needed for the variable innodb_log_block_size to be changed.

variable innodb_log_file_size¶

Version Info:	1.0.6-10 – Introduced
Command Line:	Yes
Config File:	Yes
Scope:	Global
Dynamic:	No
Type:	Numeric
Default Value:	5242880
Range:	1048576 .. 4294967295

In upstream MySQL the limit for the combined size of log files must be less than 4GB. But in Percona Server it is:

• on 32-bit systems: individual log file limit is 4 GB and total log file size limit is 4 GB, i.e. the same as in the upstream server.
• on 64-bit systems: both individual log files and total log file size are practically unlimited (the limit is 2^63 - 1 bytes which is 8+ million TB).

variable innodb_read_ahead¶

Command Line:	Yes
Config File:	Yes
Scope:	Global
Dynamic:	Yes
Variable Type:	String
Default Value:	linear
Values:	none, random (*), linear, both

This variable controls the read-ahead algorithm of InnoDB. The following values are available:

• none: Disables read-ahead
• random: If enough pages within the same extent are in the buffer pool, InnoDB will automatically fetch the remaining pages (an extent consists of 64 consecutive pages)
• linear (default): If enough pages within the same extent are accessed sequentially, InnoDB will automatically fetch the remaining pages.
• both: Enable both random and linear algorithms.

You can also control the threshold from which InnoDB will perform a read ahead request with the innodb_read_ahead_threshold variable

random is removed from InnoDB Plugin 1.0.5, XtraDB ignores it after 1.0.5.

Status Variables¶

The following information has been added to SHOW INNODB STATUS to confirm the checkpointing activity:

• The max checkpoint age
• The current checkpoint age target
• The current age of the oldest page modification which has not been flushed to disk yet.
• The current age of the last checkpoint

...
---
LOG
---
Log sequence number 0 1059494372
Log flushed up to   0 1059494372
Last checkpoint at  0 1055251010
Max checkpoint age  162361775
Checkpoint age target 104630090
Modified age        4092465
Checkpoint age      4243362
0 pending log writes, 0 pending chkp writes
...