Speeding up query with join, where, group by, and count

I have an elementary query that is taking too long to execute even on small tables (<100,000 rows):

select images.classification, count(boxes.id)
from images
join boxes on images.id = boxes.image_id
where boxes.round = 0
group by images.classification;

I have indices on boxes.round, boxes.image_id, and images.classification (only varchar in this开发者_开发知识库 query). Primary keys on boxes.id and images.id. Explain indicates that it is taking advantage of the boxes.round index. The extra's are: Using where; Using temporary; Using filesort.

Is it possible to speed up this query? How?

If it matters, server is MySQL 5.1 all with MyISAM tables.

(This question is similar to How to speed up "select count(*)" with "group by" and "where"?)

Full explain output:

mysql> explain select images.classification, count(boxes.id) from images join boxes on images.id = boxes.image_id where boxes.round = 0 group by images.classification;

|  1 | SIMPLE      | boxes  | ref    | ix_boxes_image_id,ix_boxes_round | ix_boxes_round | 5       | const                       | 64162 | Using where; Using temporary; Using filesort |
|  1 | SIMPLE      | images | eq_ref | PRIMARY                                        | PRIMARY               | 4       | vatic.boxes.image_id |     1 |                                              |

Add an index on (images.id, images.classification) - that ought to give you

• an index for the join. You already have that via the primary key, but you'll also get:
• the same index usable as a covering index
• the same index for the grouping clause

Also add an index on (boxes.image_id,boxes.round):

• index for the JOIN
• covering index for the COUNT (see below)

About the COUNT clause: if you don't have NULLs in boxes.id (assuming not), you can replace it with COUNT(boxes.image_id) so we can get some more use from the index from prev. paragraph.

Again, verify with EXPLAIN, but I'd say these steps should get you a better execution plan.

Add an index to images.id.

Would it be feasible to use an integer for images.classification instead of a varchar?