about disk io

[??????????]

hbase In a high-density random read scenario, the current bottleneck appears in 
the disk throughput, but the disk IOPS is not high, if I reduce the blocksize 
of the table, can increase iops to reduce disk throughput?