Databases Reference
In-Depth Information
Figure 13.19
STMM controller.
space constraints, we have chosen to describe only the MIMO controller in detail since
it is the algorithm that is most often used.
Determining how frequently to tune is a key consideration for a memory controller.
An OLTP workload with thousands of short-running transactions can reasonably be
tuned every few seconds, while a complex query environment may require several min-
utes or hours before a representative window of activity has occurred in order to make
informed tuning decisions. The range of reasonable tuning rates varies by orders of mag-
nitude depending on the system workload. To our knowledge, no research team or ven-
dor has yet published a technique for determining the tuning rate for memory allocation
in an RDBMS. The few publications that discuss this topic used fixed time intervals.
STMM determines the tuning interval by observing the signal-to-noise ratio in the
benefit data from the memory consumers and finding a time interval over which the sig-
nal-to-noise ratio is within 70%. The sample interval is determined by considering the
confidence of the benefit data. Using
P
measured benefit samples
benefit(i), i=1, 2, …,
P
, a sample mean,
mean_benefit
, and a sample standard deviation,
std_benefit,
are com-
puted. The desired sample interval size is then calculated using the following equation:
13.4
