Results (
Thai) 1:
[Copy]Copied!
Hotness-Based Policy (Hoff). Disabling a cold slice with a lower number of frequentlyreferenced cache lines to save leakage power would incur a lower number ofcache misses than powering off a hot slice. Thus, the Hoff policy selects the power-offvictims from the slices with the hotness (calculated by Equation (2)) less than a predefinedthreshold, Hth. With less than Hth hotness, the increase in cache miss ratewould possibly be less than Hth when the slice is powered off. Among the cold power-offvictims, the slices with lower hotness are chosen to be turned off, with the restrictionthat at most Noff slices are powered off at an epoch to avoid write bursts. A higherHth setting would turn off more cache slices to save power but also increase the performancepenalty. We analyze the impact of different Hth settings and find that an Hthequal to 7.5% represents a good tradeoff for the system we evaluated.5.2.3. Utilization+Hotness+Dirty-Aware Policy (UHDoff). As observed in Section 4, using theutilization alone to determine the power state would lose some power-off opportunitieswhen the referenced cache lines are seldom reused. The hotness characteristic canhelp to identify these seldom reused slices, and powering off these slices can furtherreduce the leakage power. Moreover, with similar utilization and hotness, the slice withfewer dirty cache lines would incur a lower writeback penalty before shutting downthe cache slice. Therefore, the UHDoff policy considers all three factors when decidingwhen and which slice should be disabled to save power, as illustrated in Figure 10and Algorithm 1. We first select the power-off victims from the slices with less thanUth utilization (lines 6 to 13). If there is no low-utilization slice, we instead select thepower-off victims from the slices with less than Hth hotness (lines 14 to 23). Amongthe power-off victims, at most Noff slices with fewer dirty cache lines are chosen to beturned off (lines 25 to 32). In order to maintain the inclusion property, at least MinOnslices stay at the power-on state to guarantee that the LLC size is larger than the upperlevelcaches. In this algorithm, we first consider turning off the low-utilization slices,then the cold slices. The reason is that the utilization fails to capture the power-offopportunity of the seldom reused slices for the workloads with large memory footprint,while the hotness metric can help to identify these less frequently accessed slices, asillustrated in Section 4. The dirty metric is then used to select the slice that incurslower writeback overhead. We analyze the impact of different threshold settings andempirically set Uth=30%, Hth=7.5%, and Noff=4.
Being translated, please wait..
