Minutesfrom09Nov2007

!ProtoGENI Project Meeting, 09 Nov 2007

Who/When

• attendees: Jay, Mike, Leigh (phone), Rob, David, Eric
• start @ 3:15 PM
• end @ 4:45 PM+ (Eric left at 4:45 PM)

GENI Engineering Conference Planning

• to be held in early March 2008
• would like to have something to show off by then!

What's Required for Minimum Functionality

• What is the minimum set of stuff that we need to do to Emulab to support a GENI-like service, from a user's perspective?
• Mike: what do users expect to see in a !ProtoGENI, so that they don't say "that's just Emulab."
  • Jay: we just say "it's !ProtoGENI." Don't care if they say it's Emulab.
• Rob: people expect to see shared nodes
  • ...even though we offer dedicated nodes already
  • Leigh: can't we get shared nodes just by putting the PlanetLab disk image on our own nodes? Yes!
• Mike: tunnels over the Internet
  • Jay: GENI is trying to have reserved stuff
  • Rob: GENI is supposed to have its own backbone, although it doesn't have this now
• Rob: get a virtual network, like they get in Emulab, but it's physically distributed over the country
  • Leigh: why not just run an overlay, and use tun/tap drivers?
• Mike: what do users expect from Emulab in this new world?
  • Jay: wireless nodes
  • Jay: emulated subnets on the edges
  • Rob: something that looks like or is a server cluster
  • Jay: shared wireless node
• Mike: what is the user interface to !ProtoGENI going to look like?
  • Jay: an ns file
  • Mike: what do we need to add to ns?
  • New specifications for tunneling? There are lots of ways to tunnel.
    • Rob: we already have a weak form of weak link typing --- tb-set-link-protocol
• Mike: are we going to resurrect "WAN assign"?
  • Jay: WAN assign is an older program (written by Chad) that runs a genetic algorithm to set up tunnels on RON nodes.
  • What good what this be? Once we have reserved pipes, assign will do a better job than WAN-assign.
  • Resolution: do not resurrect WAN assign.
• Mike: do we need any of the hierarchical naming schemes that are described in the specification?
  • No
• Mike: do we need to do authentication stuff?
  • Jay: no, not if everyone goes through our portal.
  • We can get some authentication stuff from our existing XMLRPC stuff, e.g., existing user certificates
• Jay: real layer 2 devices
  • Even if we don't get NLR, we can still get a metro area-wide network
  • Rob: it's valuable and crucial to the story to have layer 2 stuff between sites
    • do stuff with link typing at first, and something sexier later
  • Jay: what about PCs that are there?
• Jay: native Node Manager nodes
  • making it easier
• Jay: wide-area event system
  • Do we need to do anything to make the event system work in the wide area?
  • How to control wide-area nodes?
  • If we can't reboot things, they aren't going to work!
    • Rob: but we can't so easily reboot shared stuff
  • Mike: but we still use elvind in the wide area. Is this important?
    • Jay: how many days to fix this?
    • Leigh: less than a week, but more than a day or two
• Jay: buy layer 1 switches for us
  • It's good to be "one layer down,"to help us develop !ProtoGENI
  • Rob: not sure how much this will actually help us.
• Jay: virtual server stuff?
  • Mike: no, not for version 0
  • Rob: getting real switches (from Jon Turner, say) ASAP is good. It will distinguish us from other people doing GENI work.
• Jay: what support do we need to have programmable routers?
  • NetFPGA

Summary: VersionZero Functionality

• Shared nodes
  • Run our own PLC.
  • Run our own nodes with the PlanetLab image, and allow people to get slivers on them.
  • Our PLC talks to our ELC (?).
• Tunnels
• Two-level naming scheme
• Layer 2 devices, in a geographically distributed network
• Wide-area events

Action Items

• Leigh/Mike/All: read the new VINI paper by next meeting (see MinutesFrom13Nov2007)

-- Main.EricEide - 10 Nov 2007