In this BoF session, we will briefly discuss challenges face by next-generation ICT infrastructures, which motivate the operating system related activities in Huawei. By revisiting the history of operating system evolution and reviewing the hardware evolution, we will discuss how the operating system should be shaped to embrace the challenges raised by new application scenarios and the opportunities brought up by hardware evolution as well as research advances in the systems community. Specifically, we will argue that operating systems should serve applications requiring ultra lower and differentiated latency (faster), higher overall system throughput and better security and trustworthiness under a strong threat model.

Haibo Chen: Chief OS Technology Expert and Director of OS Kernel Lab, Huawei, and Professor, Shanghai Jiao Tong University
Xinwei Hu: Chief OS Architect and Director of OS Architecture and Design Department, Huawei

7:00-7:10 Welcome Speech
    (Jingren Zhou, Vice President of Alibaba)
7:10-7:20 NewSQL for bigdata analyzing in Alibaba
    (Wei Lin, Architect of Alibaba computation platform)
7:20-7:30 Challenges and Opportunities in Large-Scale Graph Processing at Alibaba
    (Zhengping Qian, Senior Staff Engineer, Alibaba)
7:30-7:40 Hyper scale data center network optimization practice at Alibaba
    (Yongfeng Liu, Staff Engineer, Alibaba)
7:40-7:50 Hardware/Software Co-Design for Resource-Efficient Datacenters
    (Yungang Bao, Professor, Institute of Computing Technology, CAS)
7:50-8:00 Discussion

In this BoF session, we aim to explore the efficient implementation of modern programming paradigms surrounding threads and events. There are myriad ways of solving problems in this space, and this BoF will encourage discussion around various challenges and solutions. To introduce the session, we will briefly present our priority- aware, multi-threaded event delivery system which is part of the latest iOS, macOS, tvOS, and watchOS.

Jeremy Andrus, Kernel Engineering Manager at Apple

Recently, AI and Machine Learning techniques are being applied to many gnarly infrastructure and systems problems with various degrees of success. While there is tremendous excitement (and hype!), formulating problems in a given domain such that a machine learning algorithm can be correctly applied, needs deep expertise, which may not be available to all. Hence the notion of democratizing AI with the necessary automation and toolsets has caught on, with the hope that domain experts can use these tools effectively with little help. In this session, we will discuss some questions about this.

David Tennenhouse (VMware)
Matei Zaharia (Stanford University)
Joseph Gonzalez (UC Berkeley)
Garth Gibson (Carnegie Mellon University and Vector Institute)

ZHIWANG Wealth is a data-driven online wealth management platform. It provides various financial products: fix-income debt, mutual fund, insurance, etc. By leveraging fundamental technologies such as big data, cloud computing and machine learning, ZHIWANG builds the customer persona profile for intelligent operation to achieve high productivity.

The fixed-income debt product sold in ZHIWANG is from credit-loan assets. To ensure the portfolio quality, the team uses knowledge-graph to store and process big data, and applies machine learning to build and optimize its risk management models.

Yun Zheng, Director of Engineering, CreditEase Big Data Innovation Center

We come to a new era of cyberspace enabled by the AI technology and Internet services built upon the Cloud. In this talk, we will present two use cases of networking stack implementation in Baidu's production environment. The first one is a new TCP algorithm tailored for Linux servers supporting our cloud CDN services.The second use case we share is the RDMA technology embedded into Baidu's PaddlePaddle AI framework, which is extensively used by image recognition, speed recognition and machine translation.

Yongxin, Chief Networking Architect and the Chairman of the Technical Committee for System Department at Baidu

Distributed block storage is the key component of building web-scale enterprise cloud. Sequential consistency for block storage is presumed by many applications, and should also be guaranteed for distributed block storage.

We designed and implemented ZBS, a high-performance distributed block storage system for hyper-converged infrastructure. It provides sequential consistency which is difficult to ensure in distributed system, and can also run with little CPU and memory consumption.

Kyle Zhang, Cofounder and CTO of SmartX

Didi Chuxing is one of the fastest growing internet companies riding the tide of mobile internet and sharing economy. Comparing with other internet service providers, DiDi has rigorous latency requirements on its ride hailing services in order to provide superior user experiences.In this seesion, we will share the unique challenges at system software layer that we face during the course of building the world’s largest ride hailing and smart transportation services at DiDi and explore collaborative opportunities to address these challenges.

Yong Wang, Engineering Director at DiDi’s Infrastructure Platform Division