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 c...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.展开更多
On behalf of the program committee co-chairs of IFIP 2011 International Conference on Network and Parallel Computing （NPC 2011）, we are very happy to organize a special section of five best paper nominees selected f...On behalf of the program committee co-chairs of IFIP 2011 International Conference on Network and Parallel Computing （NPC 2011）, we are very happy to organize a special section of five best paper nominees selected from the conference. The committee reviewed our 54 submissions over a 6-week period from mid-April to the end of May 2011,展开更多
In the past few years, we have witnessed rapid proliferation of online social network services, such as Facebook, Twitter, and Del.icio.us, which greatly facilitate the collaboration, sharing, and other kinds of inter...In the past few years, we have witnessed rapid proliferation of online social network services, such as Facebook, Twitter, and Del.icio.us, which greatly facilitate the collaboration, sharing, and other kinds of interactions among individuals. The term ＇social media＇ has thus been coined to embrace all those new collaborative services or applications and also to indicate a new ＂social＂ approach to generating and distributing Web contents. The everincreasing social network services produce large-scale social data for analyzing user behaviors, and then improving existing services. On the other hand, they also bring forth more severe information overload and require more flexible and robust computing platforms or service paradigms.展开更多
基金This work is supported in part by the National Key Research and Development Program of China under Grant No.2016YFB1000201the National Natural Science Foundation of China under Grant Nos.61420106013 and 61702480the Youth Innovation Promotion Association of Chinese Academy of Sciences and Alibaba Innovative Research(AIR)Program.
文摘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.
文摘On behalf of the program committee co-chairs of IFIP 2011 International Conference on Network and Parallel Computing （NPC 2011）, we are very happy to organize a special section of five best paper nominees selected from the conference. The committee reviewed our 54 submissions over a 6-week period from mid-April to the end of May 2011,
文摘In the past few years, we have witnessed rapid proliferation of online social network services, such as Facebook, Twitter, and Del.icio.us, which greatly facilitate the collaboration, sharing, and other kinds of interactions among individuals. The term ＇social media＇ has thus been coined to embrace all those new collaborative services or applications and also to indicate a new ＂social＂ approach to generating and distributing Web contents. The everincreasing social network services produce large-scale social data for analyzing user behaviors, and then improving existing services. On the other hand, they also bring forth more severe information overload and require more flexible and robust computing platforms or service paradigms.
