2020-05-06 SmartNIC SC Meeting Agenda/Minutes

Agenda

• Ground rules - openness of the SmartNIC wiki page 
• Proposed Initiative Review

  • Initiative Proposed by Arm: Standardisation

  • Initiative proposed by Futurewei: NOF Initiator

• Next Steps

Minutes

Ground rules - openness of the SmartNIC wiki page

• Andrea: by default the SmartNIC interim project home page is closed and restricted to the white list of members in this group plus Linaro Core, Club and LDCG members. It will become open as soon as the project launches officially with committed members. Is this ok or shall we open it up right now?
• Maciek: everything shall be open from the beginning.
• WanLi: everything shall be open.
• Donna: yes, open.
• Tomas: agreed, very hard to forward any material internally if not open.
• Andrea: great, let’s send out an official vote email and if positive we will then open the pages to the public immediately

• Grant: we would like to clarify that current requirements for the project require 1 engineer. Not 1 engineer per initiative, 1 engineer for the entire smartNIC project. 1 vote per member.
  Tomas: one individual full time engineer or FTE or other options? Grant: this is flexible. Let’s discuss in the upcoming project structure review calls or in 1-1 with Linaro if any custom requirement / proposal.
• Al: is it any Linaro Group or exclusively LDCG? Grant: exclusively LDCG.

Proposed Initiative Review

• Paul: See the template with basic info in slide 5
• Initially we publish the proposals in the front page of the SmartNIC portal, then one by one turn into an official Initiative in Jira when ready

Initiative Proposed by Arm: Standardisation

SLIDES HERE: SmartNIC Standardisation Initiative

• Grant describes the initiative he is proposing.
• Sam Fuller: there are requirements around TPM support, as these get deployed in data centers. There is a need for an integrated hw root of trust. Is this included?
• Grant: deployment of keys, etc. is all included.
• Sam: shall we have this as a separate security initiative?
• Grant: yes, it would make sense to break the security portions out. (Possibly security lifecycle)
• Graeme: this is a separate process
• Tomas: if you do have separate domains accessing the same smartNIC, you shall have the division well firewalled from each other. (Well partitioned portions from one another.) 
• Tomas: didn’t see the standardization towards motherboard re: BMC interactions. (Grant agrees and will add to the slide.) (Question on which manages which: BMC managing the machine or machine managing the BMC. Answer is probably both in different use cases. (Graeme confirms this from his perspective.)
• Grant: suggests identifying a few useful use cases and identify the related work to enable those use cases.
• Al concurs there are various use cases represented here: initial boot, etc. represented through the lifecycle of the device.
• Grant: Do we need an activity to identify the life cycle use cases?
• Al, Sam: yes
• Al: Being able to walk through these steps is important
• Tomas: separation of concern when different operators accessing the same device - will have different life cycles.
• Grant: yes, this is critical background that we shall work on - let’s create the primary life cycle use case. Any volunteer to take a first cut?
• Al and Tomas volunteer to work on this.
• Grant will start a thread on the mailing list
• Paul to create a page in confluence
• Create an initiative for Security TPM (Depends on lifecycle cases.)
• Grant asks if Sam would take the first pass at security initiative. 
• Sam will share a couple of slides on the security aspects of this.

Initiative proposed by Futurewei: NOF Initiator

SLIDES HERE: NOF Initiator

• Wan Li from Futurewei describes the SmartNIC NOF initiator initiative
  • Linux supports NOF initiator for a few years
  • VMware vSphere 7 added NOF support this March
  • Windows doesn’t have native NOF initiator
• Separates Compute & Storage (Scale compute & storage separately)
• End-to-end NOF solution requires collaboration from Server OS, Hypervisor, NIC/HBA, Storage OS and Storage drives
• Current OS/Hypervisor NOF is not universal
• NOF hardware initiators from a few vendors (Univ. of New Hampshire)
• Sam: can you tell us more about ROCE on TCP? 
• Wan Li: ROCE over TCP is very new, not sure about this yet. We believe it will become a bigger player
• Sowmini: I was having this discussion 1y ago in a different life/role, the problem is that ROCE is not standardized and there is no interest from ROCE to become standardized. Mellanox is completely upset, they are used to work with standards.
• Sam: in HPC vs datacenter environment, there is a lot of routing going on in the rack and need something like TCP
• Sowmini: yes but TCP was not meant to be offloaded like this. There is one company that is trying to corner the market here and we need to be aware of this.
• Tomas: fully subscribe this, we shall be fully open
• Tomas: would you like to see all this on dedicated links? It doesn’t play well with other ethernet workloads. Would need some allocation of bandwidth so that you can have predictable performance from the storage subsystem. The smartNIC shall have a dedicated port for this.
  • Wan Li: thanks, noted that.

Next Steps

• Next call is in one week and it is to further discuss the membership structure

Action Items

• Paul Isaac's (Deactivated) send vote request to open the SmartNIC page, meeting minutes and call recordings immediately and record on Collaborate
• Paul Isaac's (Deactivated) create SmartNIC sub-pages in Collaborate to document the life cycle and security use cases
• Grant Likely initiate email threads on smartnic@linaro.org to discuss the proposed use cases
• Al Stone (Deactivated) describe the life cycle use case with contributions from Tomas Fredberg
• Sam Fuller describe the security use case 

Recording Links