Labeled yon Neumann Architecture for Software-Defined Cloud

As cloud computing is moving forward rapidly, cloud providers have been encountering great challenges： long tail latency, low utilization, and high interference. They intend to co-locate multiple workloads on a single server to improve the resource utilization. But the co-located applications suffer from severe performance interference and long tail latency, which lead to unpredictable user experience. To meet these challenges, software-defined cloud has been proposed to facilitate tighter coordination among application, operating system and hardware. Users＇ quality of service （QoS） requirements could be propagated all the way down to the hardware with differential management mechanisms. However, there is little hardware support to maintain and guarantee users＇ QoS requirements. To this end, this paper proposes Labeled von Neumann Architecture （LvNA）, which introduces a labelling mechanism to convey more software＇s semantic information such as QoS and security to the underlying hardware. LvNA is able to correlate labels with various entities, e.g., virtual machine, process and thread, and propagate labels in the whole machine and program differentiated services based on rules. We consider LvNA to be a fundamental hardware support to the software-defined cloud.

A Case for Adaptive Resource Management in Alibaba Datacenter Using Neural Networks

Both resource efficiency and application QoS have been big concerns of datacenter operators for a long time,but remain to be irreconcilable.High resource utilization increases the risk of resource contention between co-located workload,which makes latency-critical(LC)applications suffer unpredictable,and even unacceptable performance.Plenty of prior work devotes the effort on exploiting effective mechanisms to protect the QoS of LC applications while improving resource efficiency.In this paper,we propose MAGI,a resource management runtime that leverages neural networks to monitor and further pinpoint the root cause of performance interference,and adjusts resource shares of corresponding applications to ensure the QoS of LC applications.MAGI is a practice in Alibaba datacenter to provide on-demand resource adjustment for applications using neural networks.The experimental results show that MAGI could reduce up to 87.3%performance degradation of LC application when co-located with other antagonist applications.

高通量计算的兴起（英文）

近年来,云计算、人工智能和物联网等新兴计算应用的出现,对计算机系统设计提出3个共同要求:高利用率、高吞吐量和低延迟。在这里,这些被称为"高通量计算"的要求。我们进一步提出一种新指标,称为"系统熵",用于测量计算机系统的混乱程度和不确定性。我们认为,与追求高性能和低功耗的传统计算系统的设计不同,高通量计算应致力于实现低并发性。然而,从计算机体系结构角度看,高通量计算面临两大挑战:(1)如何充分利用应用程序数据的并行和并发执行来实现高吞吐量;(2)如何获得低延迟,即便在具有高利用率的数据路径中发生严重争用的情况下。为应对这两个挑战,引入两种技术:片上数据流体系结构和标签化冯诺依曼体系结构。构建了两个可以实现高吞吐量和低延迟的原型,显著降低了系统熵。