Anticipatory IO Scheduler
Expected IO Scheduler
-------------------------
Nick Piggin <[email protected]> SEP 2003
attention! Database servers, especially those using "TCQ" disks should
Investigate performance with the ' Deadline ' IO scheduler. Any system with high
Disk performance requirements should do and in fact.
Attention! database servers, especially those using "TCQ" disks should be studied using the ' deadline ' IO scheduler
The performance. In fact, any system that uses high-performance disks should take these into account.
If you see unusual performance characteristics of your disk systems, or you
See big performance regressions versus the deadline scheduler, please email
Me. Database users don ' t bother unless you ' re willing to test a lot of patches
From me;) its a known issue.
If you see unusual performance characteristics in your disk system or you see a high-performance regression analysis and use of
If the term is scheduled, please send me an e-mail. Database users don't have to bother, unless you want to test a lot of them here
The patch.
Selecting IO schedulers
Select IO Scheduler
-----------------------
To choose IO schedulers @ boot time, use the argument ' Elevator=deadline '.
' NoOp ' and ' as ' (the default) are also available. IO Schedulers is assigned
Globally at boot time is only presently.
In order to be able to select IO Scheduler during startup, use the parameter ' Elevator=deadline '. NoOp ' and ' as ' are also
OK. Currently, the IO Scheduler is only able to be globally allocated during startup.
Tuning the Anticipatory IO Scheduler
Tuning expected Dispatch
------------------------------------
When using ' as ', the anticipatory IO Scheduler there is 5 parameters under
/sys/block/*/iosched/. All is units of milliseconds.
When using ' as ', the scheduler is expected to be under the directory/sys/block/*/iosched by 5 parameters, these 5 parameters are
milliseconds.
The parameters are:
These five parameters are:
* Read_expire
Controls how long until a request becomes "expired". It also controls the
Interval between which expired requests is served, so set to, a request
Might take anywhere < 100ms to be serviced _if_ It's the next on the
Expired list. Obviously it won ' t make the disk go faster. The result
Basically equates to the TimeSlice a single reader gets in the presence of
Other IO. 100* (Seek Time/read_expire) + 1) is very roughly the%
Streaming read efficiency your disk should get with multiple readers.
control when a request is out-of-date. At the same time he also controls which expired requests can be served by the interval,
So it is set to multiply by 50. The next request on the overdue list may be taken anywhere in less than 100 milliseconds when the
Service. Obviously, he won't make the disk faster. This result is basically equivalent to a single in the other IO
Read the program to get a time slice.
* Read_batch_expire
Controls How much time a batch of reads was given before pending writes are
Served. Higher value is more efficient. This might is set below Read_expire
If writes is to being given higher priority than reads, but reads is to be
As efficient as possible when there is no writes. Generally though, it
Should be some multiple of read_expire.
Controls how much time is given to a batch of read programs before some waiting write programs are serviced. The high values are more by efficiency.
If the write operation is given a higher priority than the read operation, this value may be set to a lower force read_expire.
* Write_expire, and
* Write_batch_expire is equivalent to the above, for writes.
* Write_expire, and
* Write_batch_expire is set for the write program, same as above.
* Antic_expire
Controls the maximum amount of time we can anticipate a good read before
Giving up. Many other factors could cause anticipation to be stopped early,
Or some processes is not being "anticipated" at all. Should be a bit higher
For big seek time devices though not a linear correspondence-most
Processes has only a few Ms ThinkTime.
Control before giving up we can expect a good read program to be big time. Any other factor will lead to the expected advance
Terminated, or some process is not expected at all. Even if not a linear counterpart, to a high-looking time
The equipment should also be set slightly higher. Most of the thinking time is just a few milliseconds.
As-iosched.txt document translation under the Documentation Directory
