Loading…
DPDK Summit China 2019 has ended

Sign up or log in to bookmark your favorites and sync them to your phone or calendar.

Monday, June 24
 

08:00 CST

Registration
Monday June 24, 2019 08:00 - 09:00 CST

09:00 CST

Opening
Speakers

Monday June 24, 2019 09:00 - 09:15 CST

09:15 CST

Archive stable high performance DPDK on modern multi-core CPU: Tao Yang & Xuekun, Intel
As Moore’s law keep advance, modern CPU included more and more cores. Although every core has dedicated execution unit and L1/L2 cache, all the cores in the same CPU share many resource such as L3 cache, memory bandwidth, power. Running DPDK application on multi core CPU, DPDK application performance will be affect by the utilization of these shared resources. In this session we will share our study on how to partition the core power, L3 cache and memory bandwidth to archive stable high performance for DPDK application.

Speakers
XH

Xuekun Hu

Platform Application Engineer, Intel
TY

Tao Yang

Platform Application Engineer, Intel


Monday June 24, 2019 09:15 - 09:40 CST

09:40 CST

DPDK IPsec: a scalable high performance library for your IPsec application: Fan Zhang & Cunming Liang, Intel
IPsec is the most commonly used security protocols in 4G-LTE, and is recognized as the method of securing control and data channels of 5G. For adapting the massive traffic encryption needs in 5G, we propose our DPDK IPsec library, a scalable, high performant, and modularized library with simple but comprehensive APIs to ease your IPsec application development. Moreover, DPDK IPsec library is designed with seamless adoption of different software and hardware security accelerators including lookaside and inline accelerators, and can work in both native and virtual platforms without application change._x000D_
In this presentation we will describe the modular architecture of the IPsec library and present an example integrations of the library into the data plane application, including the performance indicators of different use-cases.

Speakers
avatar for Fan Zhang

Fan Zhang

Front-end Engineer, 3M
Fan Zhang is a Front-end Engineer in the 3M Data Science Community and a PhD candidate who specializes in fine-grained access control performance optimization in Edge Computing with the Emerging Networks Laboratory at Walton Institute in Ireland.
CL

Cunming Liang

Architect, Intel
Cunming is architect of data plane software, spending years on high performance network I/O of general purpose CPU and its virtualization includes para-virtualization and SR-IOV. One of the recent effort is to drive NFV grade virtual network device towards higher performance, smaller... Read More →


Monday June 24, 2019 09:40 - 10:05 CST

10:05 CST

DPDK Slab Allocator and applied it to zero-copy stack: Yijun Hou, Alibaba
Currently,_x000D_
#1:DPDK application uses mempool for better performance. But mempool is for fix size buffer. Once application need different size buffer, it will need different mempool._x000D_
#2:And rte_malloc is the lower performance because it has a global lock for allocate different size buffer._x000D_
For best performance and best dpdk memory utilization, we implemented DPDK Slab Allocator. It can for different packet size and has better performance._x000D_
More important, it can match dpdk application zero copy memory requirement based on user space stack and DPDK.

Speakers
YH

Yijun Hou

Engineer, Alibaba


Monday June 24, 2019 10:05 - 10:30 CST

10:30 CST

DPDK based userspace TCP/IP stack testing: Shu Ma, Alibaba
With the rapid development of NIC, to achieve extremely low latency and high throughput, DPDK based userspace network frameworks, such as Tencent's f-stack, scylladb's seastar, and AliCloud's Luna, are playing very important roles in the high performance scenario. One of the pros of userspace TCP/IP stack is customizability. We may want to apply new congestion control algorithm or tune arguments, to gain better performance. So, an easy-to-use test framework is needed to verify the change. Packetdrill is an extremely easy and extensible TCP/IP stack testing tool, developed by Google. We made some modifications to it, to make it available for both kernel and userspace TCP/IP stack. Luna uses it to reproduce stack bugs, and verify new changes in our CI system._x000D_
_x000D_
In this proposal, I will briefly introduce this tool, covering its external usage, internal mechanism, and our recent practice. Hopefully, other developers can draw their inspiration from our practice, to ease their network stack test.

Speakers
SM

Shu Ma

Senior Software Engineer, Alibaba


Monday June 24, 2019 10:30 - 10:45 CST

10:45 CST

AM Break
Monday June 24, 2019 10:45 - 11:00 CST

11:00 CST

Developing CI/CD for NICs DPDK PMD using DTS: Ben-Shahar Belkar, Toga Networks
DTS (DPDK Test Suite, https://doc.dpdk.org/dts/gsg/) is a python based framework for performing functional and performance tests for DPDK. It is a full Open Source project, started by Intel in 2014. It supports testing nics from Intel, Mellanox and Cavium Networks. Recently the DPDK performance lab in the University of New Hampshire (https://www.iol.unh.edu/testing/hpc/dpdk) adopted DTS for performance testing as part of the CI/CD._x000D_
Rami Rosen was leading the upstreaming of Huawei Smartnic PMD, which is targeted for DPDK 19.08, and for DTS enablement for this PMD, and for using DTS as part of CI/CD for Huawei Smartnic._x000D_
This talk will give an overview of the DTS project, the current status and roadmap, advantages of DTS, the challenges for enabling DTS for Huawei Smartnic PMD, and more.

Speakers
avatar for Ben-Shahar Belkar

Ben-Shahar Belkar

NetIOLabs Software R&D Lead, Toga Networks


Monday June 24, 2019 11:00 - 11:20 CST

11:20 CST

Building DPDK Unikernels with Unikraft: Sharan Santhanam, NEC Laboratories Europe GmbH
In​ ​recent​ ​years unikernels have​ ​shown​ ​immense​ ​performance potential​ (e.g., throughout in the range of 10-40 Gb/s, boot times of only a few ms, image sizes of only hundreds of KBs).The​ ​fundamental​ ​drawback​ ​of​ ​unikernels​ ​is​ ​that​ ​they​ ​require​ ​that applications​ ​be​ ​manually​ ​ported​ ​to​ ​the​ ​underlying​ ​minimalistic​ ​OS, needing​ ​both​ ​expert​ ​work​ ​and​ ​often​ ​considerable​ ​amount of​ ​time.​
The Unikraft project provides a unikernel code base and build system that significantly simplifies the building of unikernels. In this talk, we will describe our recent port of DPDK to Unikraft, as well as present early performance numbers from running DPDK in a unikernel.

Speakers
avatar for Sharan Santhanam

Sharan Santhanam

Software Specialist, NEC Laboratories Europe GmbH



Monday June 24, 2019 11:20 - 11:45 CST

11:45 CST

Networking Products Based on DPDK Integrated with Kubernetes Container Networking Contiv: Zijiang Zhou, Jingdong Digital Technology
In order to achieve the network goals of high throughput, minimal latency and high-availability in the kubernetes cluster, we have implemented a completed network system which is based on dpdk and contiv. In our network model, the communications between container to container are direct, encapsulation and network address translation are not needed. Furthermore, we make the container network and the underlying network interconnected. For the domain name system in kubernetes, we have developed the high performance jdd-dns which is based on dpdk. For the load balance in kubernetes, we have developed the jdd-slb which is also based on dpdk. And the jdd-slb is integrated into kubernetes by self-developed slb-controller, so the kube-proxy components on every cluster node are not needed. Besides, many improvements and optimizations have been made in contiv, for example, supporting multiple bgp neighbors, supporting remote container runtime containerd and cri-o.

Speakers
ZZ

Zijiang Zhou

Senior Cloud Developer, Jingdong Digital Technology


Monday June 24, 2019 11:45 - 12:10 CST

12:10 CST

Lunch
Monday June 24, 2019 12:10 - 13:10 CST

13:10 CST

5G Core network load test system with DPDK: Xiaohua Wang, Ericsson
In order to fulfill 5G eMBB, mMTC, uRLLC usage scenarios, 5G core network system shall promise extreme high bandwidth, ultra low latency performance characteristic_x000D_
Then how to ensure, measure and verify the 5G core network system performance characteristic is a big chanllenge._x000D_
This session will introduce a load test system developed with DPDK to examine 5G core network system performance quality._x000D_
System design evolution with DPDK from bare metal to container will be touched._x000D_
Practical problems solving and performance tuning experiences on Intel NIC and Mellanox NIC including SR-IOV will be shared.

Speakers
XW

Xiaohua Wang

Principal Software Engineer, Ericsson
More than 15 years software development experience in carrier grade communication network system and multimedia delivery network system including 5GC (3GPP 5G Core Network System), MBMS (3GPP Multimedia Broadcast Multicast Service System),MDN (Media Delivery Network System), MTV (Mobile... Read More →


Monday June 24, 2019 13:10 - 13:35 CST

13:35 CST

DPDK: Accelerate Remote Rendering of Cloud Gaming: Jingjing Wu & Owen Zhang, Intel
As 4G/5G bandwidth is making high quality streaming possible, Mobile Cloud Gaming becomes a new direction in cloud. In Intel® Cloud Gaming with Remote Rendering solution, the rendering is done remotely on Intel® VCA2(2nd generation visual cloud accelerators). In it, the graphics and encoding result interaction between VM and accelerators is based on TCP stream socket, i.e. kernel build-in network stack. It will introduce additional network setting while cloud deployment, which may not be accepted by some cloud gaming providers. A no-IP based interaction between VM and accelerators would help to fix this gap._x000D_
In this talk, we will present the idea how to combine DPDK and AF_VSOCK stream socket techniques to provide a reliable transport channel and flexible resource balance between VM and accelerators. And we will also present the work we did and the performance comparison with previous solution.

Speakers
JW

Jingjing Wu

Software Engineer, Intel
OZ

Owen Zhang

Software Engineer, Intel
Zhang, Xin (Owen), Software engineer in Intel Data Center Group (DCG). Currently he works on media module driver development, and GPU container enablement for DCG customer deployment. He made one speaking for Container with Media Server Studio integration in DCG NPG Asia summit 2... Read More →


Monday June 24, 2019 13:35 - 14:00 CST

14:00 CST

Flow and HQoS DPDK Acceleration Using 100GBE Intel Programmable Acceleration Card: Rosen Xu, Intel
This presentation discussed the new 100Gbe Intel Programmable Acceleration Card Ethernet feature in DPDK, how to transit from dedicated NIC to FPGA based Acceleration NIC using Intel FPGA._x000D_
Telcos and Cloud providers are looking for the balance between performance and expansibility when building next generation data centers. Traditional solutions would have to rely either on software solutions, which is very flexible but low performance, or on dedicated silicon hardware, which has high performance but poor expansibility. To solve this challenge, this FPGA based card provides the high performance and versatility of FPGA acceleration IP offers both inline and lookaside acceleration._x000D_
We introduce a classification methodology that enables a hybrid data plane consist of DPDK and the FPGA Acceleration IP. Aside from this, we also present unified software API including silicon NIC features and Acceleration IP features.

Speakers
avatar for Rosen Xu

Rosen Xu

Senior Software Engineer, Intel
Rosen(Weihua) Xu is a senior software engineer at Intel Network Platforms Group (NPG). He has over ten years of experience in Linux Kernel, FPGA design, system virtualization and CPU acceleration.


Monday June 24, 2019 14:00 - 14:15 CST

14:15 CST

PM Break
Monday June 24, 2019 14:15 - 14:35 CST

14:35 CST

Balancing Application Performance and Portability: Tim O'Driscoll, Intel
Data plane applications typically need 2 things – maximum performance, and the ability to be deployed on a wide range of hardware platforms. Currently, it’s difficult to achieve both. Portable applications typically use virtio and run on top of a DPDK-accelerated vswitch, but are limited by the performance of the vswitch. High throughput applications typically use SR-IOV, but need to be updated to run on platforms with new NICs._x000D_
This presentation will outline the challenges faced in combining performance and portability, and will discuss planned changes to DPDK and DPDK-accelerated vswitches to help address these. These include: enhancements to DPDK-accelerated vswitches to improve performance, the introduction of AF_XDP which provides a good balance of performance and portability for some use cases, and improvements to the DPDK ethdev API to make SR-IOV-based applications more portable.

Speakers
avatar for Tim O'Driscoll

Tim O'Driscoll

Software Product Marketing, Intel
Tim is part of the Software Product Marketing team in Intel’s Network Platforms Group. His main focus is on the Data Plane Development Kit (DPDK), but he also works on other related open source projects such as OVS and FD.io.


Monday June 24, 2019 14:35 - 15:00 CST

15:00 CST

Vhost and virtio on ARMv8 performance tuning and optimization: Joyce Kong & Gaven Hu, Arm
In this session, we will talk about the performance tuning and optimization practices we have done on ARMv8, some progress has been made and the work is still ongoing._x000D_
The optimization ways we resorted to include but not limited to: Identify data/address/control dependency and remove unnecessary barriers; use as relaxed as possible memory barriers if they are a must; inline functions to save the calling cost, correct prefetching to speed up data availability, avoid cache bouncing to help multi-core scaling, and etc. We will discuss the patches in principle, comes out with real performance comparison and analysis. In the last, we will discuss about upstreaming these patches and future plan.

Speakers
avatar for Gavin Hu

Gavin Hu

Principal Software Engineer, ARM
Gavin Hu is a Principal Software Engineer at Arm working on DPDK, he ever worked for Juniper networks, Nokia Networks, Intel and Arm. He gained rich experiences and knowledge in this industry. He got his master degree in Computer Architecture in 2004 from Huazhong University of Science... Read More →
avatar for Joyce Kong

Joyce Kong

Software Engineer, Arm
Joyce Kong, software engineer working on DPDK in Arm. She got the master degree from Shanghai Jiaotong University in 2017 .



Monday June 24, 2019 15:00 - 15:25 CST

15:25 CST

Practice of ovsdpdk in Baidu: Linsi Yuan, Baidu
1. share the experience how we do the hotupgrade both for ovs-kernel and ovs-dpdk, with no extra resource cost and keeps only a short break time
2. share the experience how we optimise the performance of ovsdpdk to deal with over 10mpps per VM
3. share the experience how we integrate the ovsdpdk with kata-container

Speakers
avatar for Linsi Yuan

Linsi Yuan

Software Engineer, Baidu


Monday June 24, 2019 15:25 - 15:40 CST

15:40 CST

Integrating AF_XDP into DPDK: Xiaolong Ye, Intel
AF_XDP is a new address family that is optimized for high performance packet processing, AF_XDP sockets enable the possibility for XDP programs to redirect packets from the driver to a user space memory buffer. It can achieve more than 10 times the performance improvement compared to AF_PACKET. This presentation will talk about our work on integrating DPDK and AF_XDP by writing a dedicated DPDK Poll Mode Driver for AF_XDP that allows DPDK applications to benefit from it without any modifications. We will give a basic introduction of AF_XDP and explain why we need an AF_XDP PMD in DPDK and extra value it could bring. We will also share some implementation details such as how to achieve zero copy by external mbuf buffer, multi-queue support, busy poll support etc. Besides that, performance comparison and future optimization work will also be presented.

Speakers
XY

Xiaolong Ye

Senior Software Engineer, Intel



Monday June 24, 2019 15:40 - 15:55 CST
 
  • Timezone DPDK Summit China 2019
    Asia/Shanghai
  • Filter By Type