The higher the number of CPUs, the more sharply the overhead for the main memory synchronization rises, combined with a reduction in the cache hit rate.
The CPU time will increase as follows:
with 2 CPUs by 5 to 10%
with 4 CPUs by 10 to 15%
with 6 CPUs by 15 to 20%
with 8 CPUs by 20 to 30%
with 10 CPUs by 25 to 35%
with 12 CPUs by 30 to 40%
with 16 CPUs by 35 to 45%
Provided the load is distributed over tasks which run in parallel and are, if possible, independent of each other, the transaction rate can be considerably increased. The following improvements in the throughput rate can be achieved on /390 servers in comparison to a uniprocessor:
with 2 CPUs by a factor of 1.7 to 1.9
with 4 CPUs by a factor of 3.3 to 3.8
with 6 CPUs by a factor of 4.6 to 5.2
with 8 CPUs by a factor of 5.7 to 6.6
with 10 CPUs by a factor of 6.8 to 7.8
with 12 CPUs by a factor of 7.6 to 8.8
with 16 CPUs by a factor of 8.8 to 9.9
In order to satisfy the increased resource requirements caused by raising the number of tasks, extension of the main memory and an increase in the number of disk drives are absolutely essential.