The two-day event was held at The Clayton Hotel in Dublin.
Presentation and panel session with the Technical Board on: Who the Tech Board are, what their responsibilities are, recent issues that they've addressed, future technical priorities/challenges.
Technical BoardPresentation and panel session with the Governing Board on: Who the Governing Board are, what their responsibilities are, progress to date, future priorities/challenges.
Governing BoardDPDK bus infrastructure has been updated for last a few releases. Although these changes should not affect the user application, it worth mentioning the changes. In this talk, I will summarize the bus changes and mention from required modifications in drivers.
Ferruh Yigit (Intel)There are various kind of HW accelerators available with SoCs. Each of the accelerator may support different capabilities and interface. Many of these accelerators are programmable devices. In this talk we will discuss various ways to support such accelerators in a generic manner.
Hemant Agrawal (NXP)Hot plug is a key requirement for live migration. So far, the hot plug and fail-safe implementation is still not friendly for PCIe devices. This talk proposes to add a general uevent mechanism in DPDK which include the uevent monitor and failure handler, to make it easy for DPDK users to implement hot plug.
Jia Guo (Intel)Devices on the PCI bus are found by the bus probe function. For each device, the list of registered drivers (PMDs) is searched until one (only) is found for the device. This presentation proposes a mechanism to share a pci device between multiple PMDs. It may also be extendable to non-pci devices.
Fiona Trahe (Intel)This is talk about the current status and planned development of VMBus support for DPDK. This talk also gives an overview of how DPDK applications are enabled on Azure Accelerated Networking using the Fail-Safe, TAP and existing drivers. It will cover some of the requirements and plans for the future.
Stephen Hemminger (Microsoft)Encryption in today's networks is becoming ubiquitous. However, running crypto on general purpose CPUs is costly. In this talk we present joint work by NXP, Intel and Mellanox on offloading protocol processing to hardware providing better utilization of host CPU for packet processing.
Boris Pismenny (Mellanox), Declan Doherty (Intel), Hemant Agrawal (NXP)This presentation focuses on the new QoS Traffic Management API for Ethernet devices that was introduced by DPDK release 17.08, as well as the new QoS Traffic Metering and Policing API planned for DPDK release 17.11. We describe the API, device drives currently supporting it and software fall-back strategy using the SoftNIC PMD.
Cristian Dumitrescu (Intel), Jasvinder Singh (Intel)Service cores is a library that abstracts the platform, providing an app with a consistent environment. Service cores allows switching of SW and HW PMDs with no application threading changes. This talk introduces service-cores, and opens discussion on how to enable DPDK with service cores.
Harry van Haaren (Intel)Wireless Base Band Device (bbdev) proposal for DPDK that abstracts HW accelerators based on FPGA and/or Fixed Function Accelerators that assist with LTE Physical Layer processing. Furthermore, it decouples the application from the compute-intensive wireless functions by abstracting their optimized libraries to appear as virtual bbdev devices.
Amr Mokhtar (Intel)There are many large InfiniBand clusters in the HPC market, they too would like to gain the DPDK user space high packet rate processing advantage, in addition to the RDMA capabilities. I will present the basic InfiniBand and IPoIB differences from Ethernet, and present results from a live POC on a 20 node cluster with DPDK using IPoIB
Shahaf Shuler (Mellanox)The userspace summit is a good place to make a yearly summary of community changes and interactions. It is also important to describe how DPDK interact with other communities. The last part would be about community processes (repositories, distributed CI, bugs tracking, tooling, website, mailing lists and Linux Foundation).
Thomas Monjalon (Mellanox), Qian Xu (Intel)This session will be a panel discussion of the future direction of ABI stability & LTS/Stable releases. In particular it will look at the request for a yearly xx.11 LTS release with a 2 year duration.
John McNamara (Intel), Ian Stokes (Intel), Luca Boccassi (AT&T), Kevin Traynor (Red Hat)In the presentation we will describe VFd, a hypervisor for SRIOV NICs jointly developed by AT&T and Intel, which uses DPDK and acts as policy enforcement software allowing advanced configuration of SR-IOV capable Network Interfaces. We will provide overview of the use cases and new DPDK API’s to support them.
Alex Zelezniak (AT&T)When working in SRIOV mode, we would prefer to let majority of the traffic to pass in HW directly from/to wire to/from VF, while the OVS-DPDK application only needs to handle exception packet flows on the PF. To support this mode we want to show a new Representor Ports model of the HW switch, which can be controlled from DPDK.
Alex Rosenbaum (Mellanox)This presentation introduces a port representor framework to DPDK. The framework based around a virtual representor PMD and representor broker plugin for physical function devices, provides the infrastructure to allow SR-IOV virtual function ports to be configured, managed and monitored within a single control application.
Declan Doherty (Intel)This talk will focus on improving VNF safety with Virtio and Vhost-user backend. Maxime will first describe VNF architecture relying on Virtio/Vhost-user. Then, he will talk about IOMMU support for the Vhost-user backend. Finally, Maxime will provide benchmarks results and discuss ways to improve both performance & safety.
Maxime Coquelin (Red Hat)The packed ring layout is the next generation ring layout standard for Virtio, which is designed for high performance and still in the proposal stage. This talk will give a quick introduction to this new ring layout definitions and summary the current status, findings and benchmark results of the prototype in DPDK.
Zhihong Wang (Intel)A drive to deliver OPEX saving and performance where and when it's needed. Enter a new era of power optimized packet processing. This talk reviews new & existing dpdk extensions for policy based power control proposed in August and the associated performance benefits.
Chris MacNamara (Intel), Dave Hunt (Intel)pfSense is a open source firewall/vpn appliance, based on FreeBSD, started in 2006 with over 1M active installs. We are basing pfSense release 3.0 on FD.io's VPP, leveraging key DPDK components including cryptodev, while adding a CLI and RESTCONF layer, leveraging FRRouting and Strongswan.
Jim Thompson (Netgate)This talk is about our framework libmoon, a wrapper for DPDK that makes building DPDK prototypes simple and fast. We've used it for multiple research prototypes as well as our packet generator MoonGen (presented last year here).
Paul Emmerich (Technical University of Munich)In our presentation, we share the lesson learned from our experience using DPDK with Go in order to implement a software router Lagopus2 (https://github.com/lagopus/vsw). We'll explain how we carefully designed DPDK binding in Go to guarantee the type safeness and the performance at the same time.
Takanari Hayama (Igel)T4P4S is a P4 compiler supporting flexible re-targetability without sacrificing high performance packet processing. To achieve this goal, it is split into hardware dependent and independent components. This talk will show the architecture of T4P4S and the design decisions made to support DPDK.
Sándor Laki (ELTE Eötvös Loránd University)Our advanced Container Network Interface combines the benefits of containers with DPDK‘s ultra-low latency and fast packet processing and the results show 28x more performance with SRIOV, DPDK using Vhost-User with OVS-DPDK and VPP.
Kuralamudhan Ramakrishnan (Intel), Gary Loughnane (Intel)