Databases Reference
In-Depth Information
Table 6.1
Comparison of Shared Resource Architectures
that SMP does (shared bus, false sharing, etc.), does that really explain the dramatic dif-
ference in scalability? There is another phenomenon occurring that is more subtle. It's
loosely termed the “N
2
effect.” Consider the computational complexity of a 30
×
30
join. The computational complexity is analyzed in Table 6.2. In the case of the unipro-
cessor, a single CPU performs the join of 30
×
30 elements. In the case of the SMP, with
three CPUs working on the problem, the same 30
×
30 elements can be joined in one-
third the time. Now watch carefully what happens with the shared-nothing architecture.
