首页
  
学术报告390:保证GPU数据中心服务质量的同时提高利用率

发布日期: 2018/5/7  投稿: 周时强    部门:    浏览次数: 531   返回


 

报 告 人:陈全 Tenure-Track特别研究员  上海交通大学,计算机科学与工程系    

报告时间:05月11日(周五)13:30~15:00

报告地点:宝山校区计算机大楼1004室

邀 请 人:童维勤 教授

 

报告摘要:

  Modern private datacenters are being outfitted with accelerators to provide the significant compute requiredby emerging online services. It is well known that the diurnal user access pattern of user-facing services provides a strong incentive to co-locate applications for better accelerator utilization and efficiency, and prior work has focused on enabling co-location on multicore processors. However, interference when co-locating applications on non-preemptive accelerators is fundamentally different than contention on multi-core CPUs a-nd introduces a new set of challenges to reduce QoS violation. In this talk, I will introduce our Baymax s-ystemthat improves the accelerator utilization in private datacenters while guaranteeing that user-facing servic-es achieve the required Quality-of-Service. 

  Meanwhile, GPUs have also been adopted in public Clouds. However, performance fairness among concur-rent applications on GPU, which is critical in public multi-tenant Clouds, is minimally supported. Targeting the public Clouds, I will introduce an machine learning-based runtime system that enables the fair sharing inpublic Clouds without any prior knowledge of user programs.

 

报告人简介:

陈全博士现为上海交通大学计算机科学与工程系Tenure-Track特别研究员。主要研究方向包括计算机系统、计算机体系结构、数据中心资源管理等。陈全已发表英文专著1部、论文超过30篇,多数发表在Science(特刊)、IEEE Trans、ACM Trans等著名期刊以及ASPLOS、ISCA、IPDPS、ICS等计算机系统领域顶级会议上。其论文获系统领域顶级会议ASPLOS 2017 Highlights,为国内学者第2次获此奖项。其研究在相关领域得到广泛关注,Google Scholar引用超过1500次,单篇最高引用达1140次。陈全曾获2015年CCF优博、2016年上海市优博、2017年教育部自然科学一等奖(排名:3/3)、2017年IEEE TCSC Award for Excellent (Early Career Researcher) 等多项荣誉,并于2017年入选微软铸星计划。