一种基于共享公平和时变资源需求的公平分配策略

在云计算系统中,有效和公平地分配多种类型的资源是非常关键的,而通过资源共享的方式在云计算系统中分配计算和存储资源,是一种提高系统资源利用率的有效方式.而现有的研究多是基于用户需求的任务数无限制而且需求不会变化的前提下进行的.为了解决云计算资源共享系统中用户有多组数量有限的时变任务资源需求的资源分配问题,提出了一种基于资源共享公平概念的多资源公平分配机制.该机制根据用户不同时刻的有限任务资源需求和用户共享资源量建立规划模型,使全局累计占优资源份额向量满足字典序最优,证明了在这种机制下,用户所得分配满足4个属性:激励共享、帕累托最优、无嫉妒、可信性.进而在具体分配问题上,提出一种启发式算法,通过用户共享系数概念设计了分配策略,可以保证分配满足公平性的同时,用户不发生共享缺损.理论和实验结果表明:所提出资源分配机制在资源共享用户提出多组时变资源需求时,在保证用户资源分配公平和保证较高资源利用率方面取得了很好的效果.

基于时变资源的容器化虚拟网络映射算法

为了构建大规模容器化虚拟网络,达到搭建高保真、易编程的虚拟网络环境的目的,提出一种基于时变资源的容器化虚拟网络映射算法,将OVS和Docker容器化的虚拟网络切块,并分布式地映射到若干计算、网络、存储资源受限的物理宿主机。首先,根据虚拟网络的拓扑结构,将具有密切链路关系的虚拟网元进行层次化聚合,以减小问题规模;其次,对聚合后的虚拟网络节点进行重要性评分,采用广度优先搜索算法与贪心策略对虚拟网络切块,并映射到资源匹配的物理宿主机;最后,通过虚拟网元消耗资源的定时反馈对算法中的资源评价模型作运行时动态调整,从而使物理资源得到有效利用。实验结果表明,该算法可在多台低配X86宿主机上容纳超过1300网元的虚拟网络,且把网络抖动维持在0.1 ms以下。

基于时变计算资源的联邦学习设备选择算法

联邦学习(Federated Learning,FL)是一种新兴的分布式机器学习范式,其核心思想是用户设备以分布式的方式在本地训练模型,且无需上传原始数据,仅需将训练后的模型上传到服务器进行模型聚合。现有研究大多忽略了设备的计算资源会随着用户的使用模式而发生时序性变化,这会影响FL的训练进度。文中针对异构设备具有时变计算资源的特点,使用自回归模型对时变计算资源进行建模,并提出了一个设备选择算法。首先构造了长期训练时间约束下最小化每轮FL平均训练时间的优化问题,接着采用李雅普诺夫优化理论对其进行转化,最后求解得到设备选择算法。实验结果表明,与基线算法相比,所提算法能够在基本保证模型质量的同时缩短FL的训练时间和设备的平均等待时间。

基于时变网络资源的多速率控制系统建模与极点配置策略研究

考虑网络控制系统中网络资源的时变特性及多速率控制的优点,提出了一种在时变网络资源环境下基于增广状态空间的多速率控制系统建模方法,并给出了该多速率控制模型的极点配置策略。最后以MATLAB仿真的形式验证了其正确性。

Spatio-temporal Characteristics of Disturbance of Land Use Change on Major Ecosystem Function Zones in China

In recent years, because of increasing human activities, ecosystems have been substantially disturbed and their service functions have been greatly compromised. Based on the effect of land use changes on the major ecosystem services, we estimated the ecosystem comprehensive anthropogenic disturbance index(ECADI) and analyzed the spatio-temporal characteristics of changes in the ECADI in China from 1990 to 2010. The average ECADI of the major ecosystem function zones in China in 2010 is approximately 0.382. The ECADI of Northeast China and North China is slightly higher than that of Northwest China and Southwest China. Most zones have slight changes in the ECADI. The average increases of ECADI in the major ecosystem function zones in China from 1990 to 2000 and from 2000 to 2010 are 0.0024 and 0.0002, respectively. The increase is mainly due to reclamation and urbanization, whereas the decrease is due to the implementation of ecosystem protection policies. During the last 20 years, the ECADI of water resources conservation zones increased first, and then stopped. The ECADI of soil conservation zones increased first, and then declined. The ECADI of sandstorm prevention zones, biodiversity conservation zones and flooding mitigation zones increased continuously. Our results may provide proposals to the government regarding land use planning and ecosystem protection plans in the major ecosystem zones. The major ecosystem function zones in the western part of China have been protected effectively. However, the major ecosystem function zones in the eastern part of China require more protection in the future.

Spatial and Temporal Variability of Water Availability in the Yellow River Basin

The changes in hydrological processes in the Yellow River basin were simulated by using the Community Land Model(CLM,version 3.5),driven by historical climate data observed from 1951 to 2008.A comparison of modeled soil moisture and runoff with limited observations in the basin suggests a general drying trend in simulated soil moisture,runoff,and precipitation-evaporation balance(P-E) in most areas of the Yellow River basin during the observation period.Furthermore,annual soil moisture,runoff,and P-E averaged over the entire basin have declined by 3.3%,82.2%,and 32.1%,respectively.Significant drying trends in soil moisture appear in the upper and middle reaches of the basin,whereas a significant trend in declining surface runoff and P-E occurred in the middle reaches and the southeastern part of the upper reaches.The overall decreasing water availability is characterized by large spatial and temporal variability.

Spatio-temporal changes in biomass carbon sinks in China's forests from 1977 to 2008

Forests play a leading role in regional and global carbon (C) cycles. Detailed assessment of the temporal and spatial changes in C sinks/sources of China's forests is critical to the estimation of the national C budget and can help to constitute sustainable forest management policies for climate change. In this study, we explored the spatio-temporal changes in forest biomass C stocks in China between 1977 and 2008, using six periods of the national forest inventory data. According to the definition of the forest inventory, China's forest was categorized into three groups: forest stand, economic forest, and bamboo forest. We estimated forest biomass C stocks for each inventory period by using continuous biomass expansion factor (BEF) method for forest stands, and the mean biomass density method for economic and bamboo forests. As a result, China's forests have accumulated biomass C (i.e., biomass C sink) of 1896 Tg (1Tg=1012g) during the study period, with 1710, 108 and 78 Tg C in forest stands, and economic and bamboo forests, respectively. Annual forest biomass C sink was 70.2 Tg Ca-1 , offsetting 7.8% of the contemporary fossil CO2 emissions in the country. The results also showed that planted forests have functioned as a persistent C sink, sequestrating 818 Tg C and accounting for 47.8% of total C sink in forest stands, and that the old-, mid- and young-aged forests have sequestrated 930, 391 and 388 Tg C from 1977 to 2008. Our results suggest that China's forests have a big potential as biomass C sink in the future because of its large area of planted forests with young-aged growth and low C density.