Page Comparison

Attendance

Interim Project Committee Members

Resources

Click here to download >>> 2020 04 29-30 SmartNIC Slides

Agenda

• Ground Rules
• Roll call
• Roadmap for upcoming meetings
• Template for a Linaro hosted project
• Call for initiative proposals

Minutes

• The two separate meetings were consolidated into a single set of minutes - https://collaborate.linaro.org/legsc/meeting-minutes/2020-04-29-smartnic-sc-meeting-agenda-minutes

Action Items

•  @ name action item here //dd-mm-yyyy  
•  <end>

Recording Links

[Andrea] Today’s agenda - project structure and way of working. Next week we will discuss technical priorities and scope.

Ground Rules

This is an open call, we have not set up any NDA, feel free to share the meeting minutes, recordings and slides from this call within your company. We kindly ask not to further forward this material outside your companies.

Roll call

• Graeme Gregory, engineering manager in the Linaro Datacenter and Cloud Group - want to make smartNICs boring, as much as servers
• Andrea Gallo, responsible for membership development at Linaro
• Paul Isaacs, director of the Linaro Datacenter and Cloud Group
• Bill Fletcher, Linaro technical liaison Engineer
• Jammy Zhou, Linaro China
• Grant Likely, software architect in the architecture group at Arm
• Maciek Konstantynowicz - distinguished engineer in the Cisco Networking Engineering Group, covering all markets except hyperscalers. System integrator, architect and designer, cover FD.io and VPP, running the LF FDIO benchmarking and performance lab. The expectation is to understand the priorities, scope and working models, specifically focusing on use cases, development and deployment models for smartNICs in public cloud, private cloud and such. Want to understand the focus of degree of standardization for hw and sw for both the control plane and data plane and the performance of both.
• Tomas Fredberg - E/// also hw engineer by degree, work now with a lot of sw defined hw, cloud platforms by Ericsson for telco operators and specifically down in the hw layers and networking, customer base is telco operators who expect interoperability among different hw and sw stacks, be able to put VNF running on top of those cloud stacks. There is the CNTT task force - clear need for acceleration and multiple different components and PCIe cards. The expectation is to have multiple different smartnic vendors in this community, join this type of effort - we can get more standardization of the different APIs between the hw and sw, manage at all layers up to VMs and containers and orchestration for both apps and the infrastructure
• Gil Bloch - sw architect at Mellanox, lead the AI and HPC sw activities at Mellanox, smart NICs make a lot of sense, have focused on offload engines and network computing for the past ten years or more. Standardizing the basics will make our life easier.

Roadmap for upcoming meetings

Maciek: how to manage the info flow among the two sets of calls (US and Asia friendly time zones?), how to know who joined, what was discussed, decisions?

• Andrea: will circulate all meeting minutes, list of participants, slides and recordings.
• Maciek: will there be a wiki page and a mailing list?
  
• Andrea: yes, being set up right now, sending the links via email
• See here for the list of participants from yesterday's call

Andrea: there is a public holiday in China till Wednesday next weekJammy: people may be back on Thursday but some may extend their days off

• Grant: let's cancel next week call on Thursday and keep the one on Wednesday

Template for a Linaro hosted project

Maciek: is the requirement for one engineer or one FTE? how to measure the contribution from a member engineer, which may revoke a vote in the committee if not engaging?

• Andrea: we are asking for one engineer, there is some flexibility if this is 60% or 70% to be decided by the Project Committee.
• Maciek: what about FTE?
• Andrea: yes, this will also be decided by the Project Committee. Same for the minimum engagement to keep voting rights -  eg. vote if you have feet in the project. The ultimate goal is to ensure that all members are fair to each other.

Maciek: where is the colocation based?

• Andrea: it is Redcentric in London currently. We may move it to Cambridge in the future, if we evaluate it may be better or needed.

Maciek: what about cost, is Linaro covering the cost of cabling, electricity, etc. for such servers?

• Andrea: yes, it is included in the current proposed membership fee

Maciek: what about test equipment? stateless traffic, stateful traffic generators, hw stimulators or custom FPGA ?

• Andrea: currently servers with sw testing. Not custom hw traffic generators

Maciek: understood, you are using COTS testing. There are cases where we need to get down to nanosecond level and custom FPGA and test equipment. This can be VERY expensive.

• The project committee will decide if such level of testing shall be part of the effort

Maciek: regarding LKFT test results, do you compile from source?

• Graeme: yes, we have a fully automated set up that builds from sources
• Maciek: what is the development pipeline? Git, gerrit?
• Graeme: yes, it can listen to any github. If the smartNIC project works with Gerrit, the build system can listen to Gerrit, both internal github and gerrit servers hosted in Linaro or external github repositories

Maciek: the way it works now, Linaro builds one single instance of the kernel but in a smartNIC system we have the kernel on the host and the kernel running in the smartNIC plus the two user spaces. There are four combinations. How to manage this? plus the kernel and user space running in the VMs.

• Grant: we shall take decisions on which combinations make sense. For example mainline to be tested with mainline.
• Maciek: do you want to invest in the future or legacy? shall we drop all VMs and focus on containers only?
• Grant: this is the guidance we are looking for from the interim Project Committee and make decisions together as a project. Start with well defined use cases that can be accomplished. Let's try not to boil the ocean
• Maciek: it also depends on timelines, if we start in June 2020, we expect that anything produced gets into real production in June 2021 or 2022 or 2023? that guides us in terms of which use case or which user structure to work on.
• Grant: yes, what we are asking for is to propose what initiatives they are particularly interested in, what is high value, what they want to see done and a high level estimate of the effort required to accomplish it. We will use this guidance to decide what we will work on. We want to be very very focused in working with mainline projects. We do not want on git trees that sit somewhere but do not reach mainline. We want to minimise the barriers to be adopted.

Next steps

Next call is on Wednesday 6th May.