You are assuming that the problem is the switch to the bitmap scan. That is very unlikely.

The likely problem is that your data had to be read from disk, either because it has grown too large to fit in cache (in which case more RAM would help), or the cache was cold when you happened to run the test.

The switch to bitmap scan is likely a response to the same thing that caused the problem, the growth in data.

You could cluster the table on the index t_feature_ind, which would make it more efficient to pull out all of the data with any given feature.

You clearly stated you goal in the question

The goal is to significantly improve the performance of searches on the MyISAM table

OK let's tune the MyISAM side of things.

I have four(4) suggestions for you

SUGGESTION #1 : Reduce your RAM

First of all, look back 5 years ago and note what I said in my answer to the post What are the main differences between InnoDB and MyISAM?: The MyISAM storage engine caches only the index pages from the .MYI of a MyISAM table.

You set key_buffer_size to 2147483648, that 2G. Yet, your MyISAM indexes only total 718M. You need to reduce key_buffer_size to 768M just to free up 1.2G of RAM.

SUGGESTION #2 : Load All MyISAM indexes into RAM

Many are not aware of this, but you can load one or more indexes from a MyISAM table into RAM. It's called a dedicated keycache. Over 3.5 years ago, I wrote about this in the answer to the post Can I partially invalidate the MySQL table cache? (See SUGGESTION #1). It is also in the MySQL Documentation : See Cache Index Syntax

SUGGESTION #3 : Expand all VARCHARs into CHARs

If the MyISAM table has many VARCHAR fields, you could make VARCHAR fields behave like a CHAR with one DDL command

ALTER TABLE myisamtable ROW_FORMAT=FIXED;

This will expand the table is such a way that a the data type of a VARCHAR(32) is actually 32 bytes wide all the time, just like a CHAR(32). This can potentially double or triple the size of the MyISAM table but you can gain a 20-25% increase in read speed.

I wrote about this many times

• Mar 25, 2011 : Performance implications of MySQL VARCHAR sizes
• Aug 12, 2011: Which DBMS is good for super-fast reads and a simple data structure?
• Sep 20, 2011 : Best of MyISAM and InnoDB
• May 02, 2012 : Which mysql storage engine to choose?
• May 03, 2012 : Which is faster, InnoDB or MyISAM?

If you do this, you will need to go back and measure how big the MyISAM index has grown. Then, redo SUGGESTION #2 and create the MyISAM dedicated keycache to that new size.

SUGGESTION #4

Although MyISAM does not have a global cache of MyISAM data, each DB Connection has a local cache size by read_buffer_size (for sequential reads) and read_rnd_buffer_size (for indexed reads).

You have read_buffer_size set at 256K (262144). You did not set read_rnd_buffer_size at all, so it is set to the default size of 256K (262144). You may want to experiment with caching data but be very careful because those two MyISAM data caches are for each DB Connection.

GIVE IT A TRY !!!