Any way to force sequential reading of index blocks?

[Engel Sanchez]

Hello there. I have a large (700 million rows) table with a primary index. I/O operations that read the entire table are an order of magnitude faster when I drop the primary index. For example, if the table is:

CREATE TABLE test_table
(
value1 INT NOT NULL,
value2 INT NOT NULL,
value3 INT NOT NULL,
CONSTRAINT pk_test_table PRIMARY KEY(value1, value2, value3)
) engine = MyIsam;

Copying the entire table into another with a select into or issuing something like "select max(value3) as m from test_table" basically makes mysql read the entire table. When the primary key is dropped, reading is much faster (this is a linux 64 system, monitoring with iostat -x -m 3). It seems that when the index is not present, it reads the file sequentially and the I/O requests are merged and very efficient, whereas when the index is present, it's issuing many individual 4K random reads and efficiency drops an order of magnitude. So I imagine it's reading the index blocks not sequentially, but in index order and jumping around. Increasing the key_cache_block_size to its maximum value of 16k does make it read in larger chunks and about 3 times faster, but nowhere near as fast as the non-indexed case. (Somewhere in Google groups I read that Mysql decides the read block size based on that variable). Is there any way to force Mysql to do a sequential read of the index blocks in this case?

Thanks in advance for any pointers/help