In the last episode (Sep 30), Dathan Vance Pattishall said:
> I haven't notice a gain from increasing the key_buffer on a dedicated
> slave. Let's take 3.23.5x for instance. Since there is only 1 thread for
> replication, a Serialized committal of data, I wouldn't imagine that
> key_buffer at higher levels say around 50% of system memory would give a
> performance boost. In fact raising sort_buffer, read_buffer,
> read_rnd_buffer would be more of a win to reduce lock times so committal
> of the data happens faster.
> Is this observation wrong? Does raising key_buffer on a slave to around
> 50% of the memory available make a difference?
A better question to ask might be "what is my current index hit ratio,
and is it low enough that I need to bump key_buffer"?
Run SHOW STATUS LIKE 'key%', and if your read hit ratio
(1-key_reads/key_read_requests) is under 80% or so, you may benefit
from increasing key_buffer. If key_reads is very close to
key_blocks_used, you might want to shrink key_buffer because mysql has
never had to throw away a cached index block. Note that the more
memory you allocate to key_buffer, the less memory the OS has available
to cache the table data istelf.