Databases Reference
In-Depth Information
Figure 11-4. Comparison of linear scalability, Amdahl scalability, and the Universal Scalability Law
It is important to understand that the USL won't describe any real system perfectly,
because it is a simplified model. However, it is a good framework for understanding
why systems fail to provide equal bang for the buck as they grow. It also reveals an
important principle for building highly scalable systems: try to avoid serialization and
crosstalk within the system.
It is possible to measure a system and use regression to determine the amount of seriality
and crosstalk it exhibits. You can use this as a best-case upper bound for capacity
planning and performance forecasting estimates. You can also examine how the system
deviates from the USL model, using it as a worst-case lower bound to point out areas
where your system isn't performing as well as it should. In both cases, the USL gives
you a reference to discuss scalability. Without it, you'd look at the system and not know
what expectations you should have. A full exploration of this topic deserves its own
topic, and Dr. Gunther already wrote that, so we won't go into this further.
Another framework for understanding scalability problems is the theory of constraints,
which explains how to improve a system's throughput and efficiency by reducing de-
pendent events and statistical variations. It is explored in Eliyahu M. Goldratt's book
The Goal
(North River), which is an extended parable about a manager at a manufac-
turing facility. Although it might seem far removed from the realm of a database server,
the principles involved are the same as those in queueing theory and other aspects of
operational research.
