Current resource provisioning schemes in Internet services leave servers less than 50% utilized almost all the time. At this level of utilization, the servers' energy efficiency is substantially lower than at peak utilization. A solution to this problem could be dynamically consolidating workloads into fewer servers and turning others off. However, services typically resist doing so, because of high response times during reactivation in handling traffic spikes. Moreover, services often want the memory and/or storage of all servers to be readily available at all times. In this article, we propose a family of barely alive active low-power server states that facilitates both fast reactivation and access to memory while in a low-power state. We compare these states to previously proposed active and idle states. In particular, we investigate the impact of load bursts in each energy-saving scheme. We also evaluate the additional benefits of memory access under low-power states with a study of a search service using a cooperative main-memory cache. Finally, we propose a system that combines a barely-alive state with the off state. We find that the barely alive states can reduce service energy consumption by up to 38%, compared to an energy-oblivious system. We also find that these energy savings are consistent across a large parameter space.

Memory size has long limited large-scale applications on high-performance computing (HPC) systems. Since compute nodes frequently do not have swap space, physical memory often limits problem sizes. Increasing core counts per chip and power density constraints, which limit the number of DIMMs per node, have exacerbated this problem. Further, DRAM constitutes a significant portion of overall HPC system cost. Therefore, instead of adding more DRAM to the nodes, mechanisms to manage memory usage more efficiently -- preferably transparently -- could increase effective DRAM capacity and thus the benefit of multicore nodes for HPC systems. MPI application processes often exhibit significant data similarity. These data regions occupy multiple physical locations across the individual rank processes within a multicore node and thus offer a potential savings in memory capacity. These regions, primarily residing in heap, are dynamic, which makes them difficult to manage statically. Our novel memory allocation library, {\it SBLLmallocShort}, automatically identifies identical memory blocks and merges them into a single copy. Our implementation is transparent to the application and does not require any kernel modifications. Overall, we demonstrate that {\it SBLLmalloc} reduces the memory footprint of a range of MPI applications by $32.03\%$ on average and up to $60.87\%$. Further, {\it SBLLmalloc} supports problem sizes for IRS over $21.36\%$ larger than using standard memory management techniques, thus significantly increasing effective system size. Similarly, {\it SBLLmalloc} requires $43.75\%$ fewer nodes than standard memory management techniques to solve an AMG problem.

Local thermal hot-spots in microprocessors lead to worst-case provisioning of global cooling resources, especially in large-scale systems where cooling power can be 50~100% of IT power. Further, the efficiency of cooling solutions degrade non-linearly with supply temperature. Recent advances in active cooling techniques have shown on-chip thermoelectric coolers (TECs) to be very efficient at selectively eliminating small hot-spots. Applying current to a superlattice TEC-film that is deposited between silicon and the heat spreader results in a Peltier effect, which spreads the heat and lowers the temperature of the hot-spot significantly and improves chip reliability. In this paper, we propose that hot-spot mitigation using thermoelectric coolers can be used as a power management mechanism to allow global coolers to be provisioned for a better worst case temperature leading to substantial savings in cooling power.

Parallelism is the key to continued performance scaling in modern microprocessors. Yet we observe that this parallelism can often contain a surprising amount of instruction redundancy. We propose to exploit this redundancy to improve performance and decrease energy consumption. We propose a multi-threading micro-architecture, Minimal Multi-Threading (MMT), that leverages register renaming and the instruction window to combine the fetch and execution of identical instructions between threads in SPMD applications. While many techniques exploit intra-thread similarities by detecting when a later instruction may use an earlier result, MMT exploits inter-thread similarities by, whenever possible, fetching instructions from different threads together and only splitting them if the computation is unique. With two threads, our design achieves a speedup of 1.15 (geometric mean) over a two-thread traditional SMT with a trace cache. With four threads, our design achieves a speedup of 1.25 (geometric mean) over a traditional SMT processor with four-threads and a trace cache. These correspond to speedups of 1.5 and 1.84 over a traditional out-of-order processor. Moreover, our performance increases in most applications with no power increase because the increase in overhead is countered with a decrease in cache accesses, leading to a decrease in energy consumption for all applications.

While microprocessor designers turn to multicore architectures to sustain performance expectations, the dramatic increase in parallelism of such architectures will put substantial demands on off-chip bandwidth and make the memory wall more significant than ever. This paper demonstrates that one profitable application of multicore processors is the execution of many similar instantiations of the same program. We identify that this model of execution is used in several practical scenarios and term it as "multi-execution." Often, each such instance utilizes very similar data. In conventional cache hierarchies, each instance would cache its own data independently. We propose the Mergeable cache architecture that detects data similarities and merges cache blocks, resulting in substantial savings in cache storage requirements. This leads to reductions in off-chip memory accesses and overall power usage, and increases in application performance. We present cycle-accurate simulation results of 8 benchmarks (6 from SPEC2000) to demonstrate that our technique provides a scalable solution and leads to significant speedups due to reductions in main memory accesses. For 8 cores running 8 similar executions of the same application and sharing an exclusive 4-MB, 8-way L2 cache, the Mergeable cache shows a speedup in execution by 2.5x on average (ranging from 0.93x to 6.92x), while posing an overhead of only 4.28% on cache area and 5.21% on power when it is used.

Nanoscale technology promises dramatically increased device density, but also decreased reliability. With bit error rates projected to be as high as 10%, designing a usable nanoscale memory system poses a significant challenge. In particular, we need to bootstrap a sea of unreliable bits into contiguous address ranges which are preferably as large as 4K-byte virtual memory pages. We accomplish this bootstrapping through a combination of dynamic error correction codes within 32-bit blocks and a static defect map which tracks usability of these blocks. The key insight is that statically-determined defect locations can be much more powerful than dynamically correcting for unknown locations, but that defect maps are only practical at a coarse granularity. Using a combination of BCH error correction codes and a Bloom-Filter-based defect map, we achieve a memory efficiency of 60% and 13% for 4K-byte pages at 1% and 10% bit-error rates, respectively.