A technique is disclosed for managing a workload distributed across
multiple data processing systems to enhance shared resource access to meet
a common performance standard. The technique includes on at least one
system, measuring performance of the work units on the system to create
local performance data, and on at least some of the systems sending the
local performance data to at least one other system of the multiple data
processing systems. The method further includes on at least one of the
systems, receiving the performance data from the sending systems to create
remote performance data, and adjusting at least one control parameter for
accessing shared resources in response to the local and remote performance
data to modify the performance of the work units distributed across the
data processing systems to achieve the common performance standard. A
dynamic resource clustering process is also employed to enhance the shared
resource management.