I don't know how to prevent it from crashing, but one thing that I did with my large tables was partition them by date.
Here, I partition the rows by date, twice a month. As long as your query uses the partitioned column, it will only search the partitions containing the key. It will not do a full table scan.
CREATE TABLE `identity` (
`Reference` int(9) unsigned NOT NULL AUTO_INCREMENT,
...
`Reg_Date` datetime NOT NULL DEFAULT '0000-00-00 00:00:00',
PRIMARY KEY (`Reference`),
KEY `Reg_Date` (`Reg_Date`)
) ENGINE=InnoDB AUTO_INCREMENT=28424336 DEFAULT CHARSET=latin1
PARTITION BY RANGE COLUMNS (Reg_Date) (
PARTITION p20140201 VALUES LESS THAN ('2014-02-01'),
PARTITION p20140214 VALUES LESS THAN ('2014-02-14'),
PARTITION p20140301 VALUES LESS THAN ('2014-03-01'),
PARTITION p20140315 VALUES LESS THAN ('2014-03-15'),
PARTITION p20140715 VALUES LESS THAN (MAXVALUE)
);
So basically, you just do a dump of the table, create it with partitions and then import the data into it.