Can you please post your query? I also need to know your table type as
different settings effect different table types?
You are right that a SELECT COUNT(*) WHERE field = 'value' should hit
the index, but does depend on your query.
You might also try EXPLAIN before your query, which will show the
approximate number of rows you are hitting. Look at SHOW TABLE STATUS
which will give an idea of the size of the rows and indexs.
Marry one to the other and it will give an idea of the cache settings to
get the query into cache. But still depends a lot on the table type!
Wm Mussatto wrote:
> On Fri, April 11, 2008 06:47, Ben Clewett wrote:
>> Are you using MyIsam or InnoDB? Or something else?
>> In either case the speed to get a COUNT() is largely down to the speed
>> if your disks and size of disk caching. A COUNT() forces the system to
>> read every row in order to count them, and any large table is probably
>> larger than your caches.
>> In some ways this is not so important, since it is unusual for a query
>> to want to read every row of a table. (I have 250GB tables which have
>> excellent performance but would take minutes to count every row :)
>> It might be better to consider the type of queries you will be using,
>> and the type of table, and optimise for that...
>> Victor Danilchenko wrote:
>>> I am trying to optimize our DB server. We have one table which has
>>> 1.3M entries, and the keys are GUIDs (so the key space is large).
>>> However, I have it all indexed. The performance was iffy, though, so I
>>> increased memory allocation, and the searches on the indexed fields seem
>>> to be OK now. Still, a simple count(field) on this table still takes,
>>> like, 2 minutes! I am guessing i am missing something obvious, but I
>>> read through a few MySQL optimization guides, and it seems like i am
>>> covering my bases.
>>> Here is the relevant slice of my my.cnf:
>>> # * Fine Tuning
>>> key_buffer = 256M
>>> max_allowed_packet = 64M
>>> thread_stack = 512K
>>> thread_cache_size = 32
>>> # * Query Cache Configuration
>>> query_cache_limit = 32M
>>> query_cache_size = 256M
>>> query_cache_type = 1
>>> table_cache = 512
>>> sort_buffer_size = 32M
>>> I am running Ubuntu 7.04, with 1GB of RAM. The relevant fields are
>>> all indexed, but I can't imagine why a simple count() would take so
>>> long, when the actual query by value on the same field is effectively
>>> instant (after my cache setting expansion).
>>> Does anyone have an idea of what I am missing? Also, if you think
>>> any of the above settings seem wrong for a server with 1GB of RAM,
>>> please let me know.
> If the field you are counting is the first field in an index I would think
> it would go much faster (system will just use the index). If some of your
> counts are fast and some are slow (you said iffy) that might explain the
> William R. Mussatto
> Systems Engineer