Energy efficiency and energy-proportional computing have become a central focus in modern supercomputers. These supercomputers should provide high throughput per unit of power to be sustainable in terms of operating c...Energy efficiency and energy-proportional computing have become a central focus in modern supercomputers. These supercomputers should provide high throughput per unit of power to be sustainable in terms of operating cost and failure rates. In this paper, a power-bounded strategy is proposed that maximizes parallel application performance under a given power constraint. The strategy dynamically allocates power to core, uncore, and memory power domains within a node to maximize performance under a given power budget. Experiments on a 20-core Haswell-EP platform for a real-world parallel application GAMESS demonstrate that the proposed strategy delivers performance within 4% of the best possible performance for as much as 25% reduction in the minimum power budget required for maximum performance.展开更多
Antenna array gain is a relative measure of performance defined differently in various literature. Most definitions of gain are not power consistent, and thus cannot be used directly in link budget analysis. In this s...Antenna array gain is a relative measure of performance defined differently in various literature. Most definitions of gain are not power consistent, and thus cannot be used directly in link budget analysis. In this short paper, we present a power correction factor for common definitions of power gain of antenna arrays that allows them to be used in standard link budget calculations.展开更多
联邦算力物联网(IoT,Internet of things)旨在通过联邦学习深度融合算力与物联网资源,从而实现对泛在离散部署的海量物联网数据和异构资源的高效利用。为了应对联邦算力物联网中模型反演和梯度泄露等新兴隐私攻击威胁,学术界和产业界对...联邦算力物联网(IoT,Internet of things)旨在通过联邦学习深度融合算力与物联网资源,从而实现对泛在离散部署的海量物联网数据和异构资源的高效利用。为了应对联邦算力物联网中模型反演和梯度泄露等新兴隐私攻击威胁,学术界和产业界对差分隐私(DP,differential privacy)这一高效的隐私保护技术进行了广泛研究和应用。然而,现有差分隐私技术在设定隐私预算时,未考虑本地算力节点的数据特征和隐私预算分配公平性的问题,造成了严重的模型精度损失。因此,提出了一种面向联邦算力物联网的隐私预算自适应优化方案——基于克拉美罗下界差分隐私的联邦学习(FedCDP,federated learning based on Cramér-Rao lower bound differential privacy)。首先,基于克拉美罗下界理论分析边缘算力节点的隐私预算估计值,实现自适应隐私预算规划;其次,通过计算边缘算力节点的上传模型与算力聚合服务器的聚合模型之间的相似度和隐私预算占比,分析得到每个节点的全局贡献度,进一步联合隐私预算估计值公平实时地优化隐私预算设定。理论分析证明了该方案可确保本地模型严格遵守ε-差分隐私,并保证全局模型收敛。基于多个公开数据集上的实验结果表明,在满足相同隐私保护需求的前提下,该方案将全局模型精确度最多提升了10.19%。展开更多
文摘Energy efficiency and energy-proportional computing have become a central focus in modern supercomputers. These supercomputers should provide high throughput per unit of power to be sustainable in terms of operating cost and failure rates. In this paper, a power-bounded strategy is proposed that maximizes parallel application performance under a given power constraint. The strategy dynamically allocates power to core, uncore, and memory power domains within a node to maximize performance under a given power budget. Experiments on a 20-core Haswell-EP platform for a real-world parallel application GAMESS demonstrate that the proposed strategy delivers performance within 4% of the best possible performance for as much as 25% reduction in the minimum power budget required for maximum performance.
文摘Antenna array gain is a relative measure of performance defined differently in various literature. Most definitions of gain are not power consistent, and thus cannot be used directly in link budget analysis. In this short paper, we present a power correction factor for common definitions of power gain of antenna arrays that allows them to be used in standard link budget calculations.
文摘联邦算力物联网(IoT,Internet of things)旨在通过联邦学习深度融合算力与物联网资源,从而实现对泛在离散部署的海量物联网数据和异构资源的高效利用。为了应对联邦算力物联网中模型反演和梯度泄露等新兴隐私攻击威胁,学术界和产业界对差分隐私(DP,differential privacy)这一高效的隐私保护技术进行了广泛研究和应用。然而,现有差分隐私技术在设定隐私预算时,未考虑本地算力节点的数据特征和隐私预算分配公平性的问题,造成了严重的模型精度损失。因此,提出了一种面向联邦算力物联网的隐私预算自适应优化方案——基于克拉美罗下界差分隐私的联邦学习(FedCDP,federated learning based on Cramér-Rao lower bound differential privacy)。首先,基于克拉美罗下界理论分析边缘算力节点的隐私预算估计值,实现自适应隐私预算规划;其次,通过计算边缘算力节点的上传模型与算力聚合服务器的聚合模型之间的相似度和隐私预算占比,分析得到每个节点的全局贡献度,进一步联合隐私预算估计值公平实时地优化隐私预算设定。理论分析证明了该方案可确保本地模型严格遵守ε-差分隐私,并保证全局模型收敛。基于多个公开数据集上的实验结果表明,在满足相同隐私保护需求的前提下,该方案将全局模型精确度最多提升了10.19%。
