Networking/Workshop-2014-07

OCP Networking Project workshop - UNH, July 21, 2014

• Overall workshop page

Agenda/Presentations

• 11:30-11:45 How to Get Involved & Spec Submission Process - Amber Grane (OCP)
• 11:45-12:00 Update on spec contributions to incubation committee
  • [http:// Broadcom update (coming)] - Prem Jonnalagadda
  • Accton update - Jeff Catlin
  • Mellanox update - Amit Katz
  • Facebook Wedge/FBOSS update - Omar Baldonado
• 12:00-12:15 ONIE Device Status - Carlos Cardenas(UTSA)/Curt Brune (Cumulus)
• 12:15-12:30 Performance Testing Proposal Overview and proposal doc - Amit Katz (Mellanox)
• (12:30-1:30 lunch)
• 1:30-2:30 HAL discussion - Azure Cloud Switch - CJ Williams (Microsoft); Switching Abstractions - Sanjay (Dell)
• (2:30-3:00 break)
• 3:00-3:30 Optics Certification Proposal - Craig Thompson (Finisar)
• 3:30-4:00 OOB mgmt/IPMI switch reqts - Jeff Catlin (Accton)/Aurelie Fonteny (Cumulus)
• 4:00-4:30 Dell 132x40 Switch Contribution Overview - Subi Krisnhamurthy (Dell)
• 4:30-5:00 Alpha Networks 32x40 Switch Contribution Overview and PDF specification - Robert Chu (Alpha)
• 5:00-6:00 UNH lab tour

Minutes

2014-07-21 OCP Networking Minutes

Amber (OCP-F): How to Get Involved With OCP and Contribute Specs

• Getting Involved
• Spec Submission Process
• OCP Networking project wiki page
• Any questions? Contact Amber via email or IRC on Freenode (#ocp-networking) or contact Omar or Carlos
• Note on the availability of accompanying files (gerbers, bom, schematics, ...)
  • Usually when contributing, just the spec file first to the community
  • All the accompanying files are made available privately to project leads and incubation committee
  • Once entire package is accepted by the incubation committee, then the entire package (spec + files) will be made public

Update on Spec Contributions

• Prem Jonnagaladda (Broadcom), Terry Whalen (Interface Masters)
  • Incubation committee asked for:
    • new functional test plan
    • new verification test plan
    • software to be open-sourced (mgmt firmware) (license TBD)
  • New systems will be coming soon as separate specs

• Amit Katz (Mellanox)
  • Lots of things Mellanox is involved in within OCP
  • Need to follow-up on the performance testing and licensing for their contribution

• Jeff Catlin (Accton)
  • All the supporting files are available as of last Friday (July 18) - for both the rack adapter and the switch
  • rack adapter: all files uploaded (restrictive license)
  • AS5712-54x: all files uploaded (permissive license)
  • awaiting on IC for review of updated files

• Omar Baldonado (Facebook)
  • Update on Wedge/FBOSS - not **yet** contributed to OCP, but we will
  • FBOSS won't be contributed as a large monolithic NOS, but rather central libraries/pieces that people can build upon.

Carlos Cardenas (UTSA), Curt Brune and Matt Peterson (Cumulus Networks), Amit (Mellanox): Testing Update

• Reviewed ONIE certification status
  • Make USB boot optional
  • Reviewed current vendor contributions to ONIE
• Introduced HW testing discussion
  • Amit reviewed different aspects of performance testing
  • => Will be circulating a doc for this and review at next monthly project call
  • Question - what is the software layer that we'll use to do this HW testing? Minimally, some amount of programming of the ASIC has to happen to test the HW, but we don't need full protocol stack implementations (e.g., a simple snake test)
    • Corollary to server HW testing - there is a "dummy OS" that gets loaded just to run the HW.
    • Should that be provided by the individual vendors? and we specify the minimal HW programming that needs to happen?

CJ Williams (Microsoft) and Sanjay Sane (Dell): HAL Discussion

• Make switch lifecyle == server lifecycle - and their servers don't run Linux (surprise!)
• Facebook/Microsoft have been sharing designs of how we've built software on top of the switches
• Software space from provider's perspective
• Defined a higher-layer API for control plane services and a lower-layer switch abstraction interface
• Data-centric API that is extensible
• Could we just use netlink as a standard?
• Is there an 80% we can just agree on?
• Can this include the Platform management stuff (power, temperature sensors) that Mellanox/Big Switch have worked on before
• A lot of discussion with Dell, Mellanox, Broadcom, Microsoft, Facebook - will have a follow-up meeting to see how to move this forward.
  • => Potentially report back during project call?

Craig Thompson (Finisar): Optics

• Optics Certification
  • ensure functionality with no statement on quality
  • Should work independently of the actual optical spec chosen
• Pluggable optical module compliance spec
• Improved toolset for accessing, monitoring, and controlling optical modules - better ethtool?
• possible others as well: high density optics, new optical layer
• => Review during project call?

Jeff (Accton) and Aurelie (Cumulus Networks): OOB/IPMI switch requirements

• minimum hardware requirements (need luggage tag like from earlier discussion)
• HW discusion
  • ONIE, automation
  • Price point from Apple/Fidelity - less than $500 for entire switch (HW+SW)
  • storage - logging + 2 x images
  • Question - do we need more than 100Mb interfaces? Mostly no, but some folks said they need 1Gb or 10Gb as they connect to other switches that don't support 100Mb.

• software requirements
  • ONIE
  • Basic L2, L3, Management, LLDP
  • scripting and advanced configuration management
• => Follow-up? Next meeting?

Subi Krisnhamurthy (Dell): switch contribution

• Z9500: 3U, 132x40GB
• Dell is considering contributing this but wanted to hear from the community on whether this is a good contribution
• Lot of discussion :)
• This feels like a big switch - 17 T2s + 5 mgmt CPUs?
• Complication for software, e.g., how does ONIE map to this? 5 ONIE images? 17 ONIE images? something in between?
• Distinction: (a) are you optimizing the physical power/cable requirements while still keeping it disaggregated/modular, or (b) are you trying to present a single "black box"
  • like (b) if it is possible
• What are the failure domains, how is better for power, cost?
• => Follow-up - Dell will come back and present failure domains and better on power/cost than just doing this with smaller boxes (like 32x40s)

Robert Chu (Alpha Networks): switch contribution

• 32x40GB, 1280Gb
• Ready to contribute spec and make it available
• => Review at next project call

Awesome tour of UNH IOL lab!

• Lots of undergrads (and even high-school seniors) getting some amazing training!

Administrative

• Next project call - August 11, 10-11 AM pacific (see wiki). Potential agenda/follow-ups from workshop
  • Review Alpha spec
  • Review HW perf testing spec
  • Review optics testing
  • HAL status report?
  • OOB switch status report?

• Next major OCP summit - October 30-31 in Paris (quick straw poll, about 1/10 of the room said they could make Paris at that time)
  • Need to see if there is enough attendance for this and what are options are...