How do I prevent system hangs due to excessive memory slab usage?

Using the default CFQ scheduler in Red Hat Enterprise Linux 4 can sometimes result in excessive slab usage or slow I/O, if using a 32bit block device.  The problem exists because the nr_requests for the block device is set too high and exhausts the memory buffers too quickly.  This problem does not exist under Red Hat Enterprise Linux 5.x

To resolve this issue, please make the...