| Here's the proof! Our test case has eight
threads which call new and delete in equal proportion and is run first on
one CPU, then on two CPUs and then on four CPUs. We compiled with Visual
C++ 6, ran it on Windows2000 Advanced Server and used Performance Monitor
to log and evaluate the results.
Note the three yellow spikes that occur when the SmartHeap for SMP test
is running, but that do not occur when the compiler library is used. These
yellow spikes represent jumps in user-mode CPU utilization (i.e. app code
not system code) when the SmartHeap for SMP version is run. The compiler
library, on the other hand, shows near constant CPU utilization regardless
of the number of processors available. In other words, only SmartHeap
for SMP effectively utilizes the additional CPU capacity.
Second, note the values for the red line. This line represents the number
of context-switches per second. For the compiler library, the red line
is off the chart as a result of the zillions of context switches occurring.
With SmartHeap for SMP the red line is near zero because it has been optimized
to reduce heap contention and the associated context switches.
The blue lines are total processor time and the green lines are privileged
CPU time (i.e., system calls) -- so green plus yellow equals blue. Note
that during the SmartHeap for SMP tests, the yellow lines overlay exactly
on the blue lines, meaning there were very few context switches and almost
no time spent in system calls.
Evaluation units are available. Please contact us for eval form and pricing information.