Generate a lot of data (large documents and many indexes) that doesn't fit in cache, then perform random reads.

Should demonstrate disk exhaustion, where the disk queue length, utilization etc are all high.