Copyright by William Lloyd Bircher 2010 - The Laboratory for ...

of Quad-Core AMD processors, this is the dominant effect. When an active core
performs a cache probe of an idle core, latency is increased compared to probing an
active core. The performance loss can be significant for memory-bound (cache probe-
intensive) workloads. Direct performance effects are due to the current operating
frequency of an active core. The effect tends to be less compared to indirect, since
operating systems are reasonably effective at matching current operating frequency to
performance demand. These effects are illustrated in Figure 6.1.
Two extremes of workloads are presented: the compute-bound crafty and the memory-
bound equake. For each workload, two cases are presented: fixed and normal scheduling.
Fixed scheduling isolates indirect performance loss by eliminating the effect of OS
frequency scheduling and thread migration. This is accomplished by forcing the
software thread to a particular core for the duration of the experiment. In this case, the
thread runs always run at the maximum frequency. The idle cores always run at the
minimum frequency. As a result, crafty achieves 100 percent of the performance of
processor that does not use dynamic power management. In contrast, the memory-bound
equake shows significant performance loss due to the reduced performance of idle cores.
Direct performance loss is shown in the dark solid and light solid lines, which utilize OS
scheduling of frequency and threads. Because direct performance losses are caused by
suboptimal frequency in active cores, the compute-bound crafty shows a significant
performance loss. The memory-bound equake actually shows a performance
106