System Energy Efficiency Lab |
 |
 |
Home | |
People | |
Research | |
Publications | |
Sponsors | |
Contacts |
|
Proactive Thermal Management Conventional thermal management techniques are reactive in nature; that is, they take action after temperatures reach a predetermined threshold value. Such approaches do not always minimize and balance the temperature on the chip, and furthermore, they control temperature at a noticeable performance cost.
We investigate how to use an autoregressive moving average (ARMA)
predictor for forecasting future temperatures, and we propose a
proactive thread allocation technique for multiprocessor systems. When
implemented in the Solaris kernel on an UltraSPARC T1 chip, our
proactive technique achieved 60% reduction in hot spot occurrences, 80%
reduction in spatial gradients and 75% reduction in thermal cycles on
average in comparison to reactive management.
We also introduce a new thermal
predictor that utilizes the band-limited property of the temperature
frequency spectrum. The big advantage of this predictor is that it does
not require training phase. We leverage this novel predictor and
workload characterization to develop a new proactive thermal management
technique. The experimental results show that applying
this approach
reduces the processor average temperature, hottest core temperature,
product MTTF and performance by 6 C, 8 C, 41% and 72%
respectively.  
For dynamic thermal management, we introduce scheduling policies at the OS-level with negligible performance overhead. We design a novel adaptive policy, which adjusts the probability value of receiving workloads for each core. Our policy reduces the frequency of high-magnitude thermal cycles and spatial gradients by around 50% and 90%, respectively, in comparison to state-of-the-art schedulers. Reactive thermal management strategies, such as thread migration, can be combined with this scheduling policy to further reduce hot spots, temperature variations, and the associated performance cost. More Information:
|