maria-developers team mailing list archive

[Daniel Bartholomew <dbart@xxxxxxxxxxxx>]

Hello everyone,

I'm working on the documentation for the Segmented Key Cache, one of
the new features in MariaDB 5.2.

The documentation page will have "about", "syntax", and "example"
sections.

I've based the "About" section on the High-Level Description from
the worklog ( http://askmonty.org/worklog/Server-Sprint/?tid=85 ) and
from the description in the mysys/mf_keycache.c file in the source.

Here is my first draft. I've changed the wording a bit to help
it flow better and I've changed "partition" to "segment" in keeping with
the official name.

----------------------------------------------------------------------

== About Segmented Key Cache ==

A segmented key cache is a collection of structures for regular MyISAM
key caches called key cache segments. Segmented key caches mitigate one
of the major problem of the simple key cache: thread contention for key
cache lock (mutex). With regular key caches, every call of a key cache
interface function must acquire this lock. Threads compete for this
lock even when the file and pages they have acquired shared locks for
are in the key cache buffers.

When working with a segmented key cache any key cache interface
function that needs only one page has to acquire the key cache lock
only for the segment the page is assigned to. This makes the chances
for threads not having to compete for the same key cache lock better.

Any page from a file can be placed into a buffer of only one segment.
The number of the segment is calculated from the file number and the
position of the page in the file, and it's always the same for the
page. Pages are evenly distributed among segments.

The idea and the original code of the segmented key cache was provided
by Fredrik Nylander from Stardoll.com. The code was extensively
reworked, improved, and eventually merged into MariaDB by Igor Babaev
from Monty Program.
----------------------------------------------------------------------

If there are any factual errors with the above, let me know.
Improvements and suggestions are also welcome.

One thing I'm not happy with is this sentence:
    "Threads compete for this lock even when the file and pages they
    have acquired shared locks for are in the key cache buffers."

The original sentence from the worklog reads:
    So threads compete for this lock even in the case when they have
    acquired shared locks for the file and pages they want read from
    are in the key cache buffers.

... which is very confusing to me. If anyone has a better or more
accurate rewording, please send it my way.

Also, there's a note in the source code (lines 5018-5023 of
mysys/mf_keycache.c) which states:
    Unfortunately if we use a partitioned key cache with N partitions
    for B-tree indexes we can't say that the chances becomes N times
    less. The fact is that any index lookup operation requires reading
    from the root page that, for any index, is always ascribed to the
    same partition. To resolve this problem we should have employed
    more sophisticated mechanisms of working with root pages.

Do any of you have any opinions on whether or not this should be
mentioned somewhere in the documentation? The only issue I have with
mentioning it is that if I do, the first question that comes to my mind
is: "If it is not N-times less, how much less is it?" I don't have an
answer to that question (and math isn't one of my strengths).

Next up will be the "Syntax" section.

Thanks.

-- 
Daniel Bartholomew
Monty Program - http://askmonty.org