Trading Off Cache Capacity for Low-Voltage Operation

Two proposed techniques let microprocessors operate at low voltages despite high memory-cell failure rates. They identify and disable defective portions of the cache at two granularities: individual words or pairs of bits. Both techniques use the entire cache during high-voltage operation while sacrificing cache capacity during low-voltage operation to reduce the minimum voltage below 500 mV.     

Addressing workload variability in architectural simulations

The inherent variability of multithreaded commercial workloads can lead to incorrect results in architectural simulation studies. Although most architectural simulation studies ignore space variability's effects, our results demonstrate that space variability has serious implications for architectural simulation studies using multithreaded workloads. The standard solution - running long enough - does not easily apply to simulation because of its enormous slowdown. To address this problem, we propose a simulation methodology combining multiple simulations with standard statistical techniques, such as confidence intervals and hypothesis testing. This methodology greatly decreases the probability of drawing incorrect conclusions, and permits reasonable simulation times given sufficient simulation hosts.     

IPC Considered Harmful for Multiprocessor Workloads

Many architectural simulation studies use instructions per cycle (IPC) to analyze performance. For multithreaded programs running on multiprocessor systems, however, IPC often inaccurately reflects performance and leads to incorrect or misleading conclusions. Work-related metrics, such as time per transaction, are the most accurate and reliable way to estimate multiprocessor workload performance.     

Simulating a $2M commercial server on a $2K PC

As dependence on database management systems and Web servers increases, so does the need for them to run reliably and efficiently-goals that rigorous simulations can help achieve. Execution-driven simulation models system hardware. These simulations capture actual program behavior and detailed system interactions. The authors have developed a simulation methodology that uses multiple simulations, pays careful attention to the effects of scaling on workload behavior, and extends Virtutech AB's Simics full system functional simulator with detailed timing models. The Wisconsin Commercial Workload Suite contains scaled and tuned benchmarks for multiprocessor servers, enabling full-system simulations to run on the PCs that are routinely available to researchers.