Minutesfrom06Nov2007
!ProtoGENI Project Meeting, 06 Nov 2007
!ProtoGENI Project Meeting, 06 Nov 2007
Who/When
- attendees: Jay, Mike, Rob, David, Leigh (phone), Eric
- start @ 11:10 AM
- end @ 12:40 PM
Fork or Don't Fork the Emulab Code Base?
- Leigh and Mike argue that we should [not] fork the code base
- must have users for direction, testing, debugging
- one code base is easier to keep up-to-date than two
- we don't have people who are dedicated to the GENI project; all our people must work on GENI and production Emulab concurrently
- 90% code/functionality overlap suggests not to fork; we would
need to copy bug fixes back and forth
- Rob: we currently commit several bug fixes per week, mostly committed by Leigh
- Rob: look at what's happened to PlanetLab
- Jay argues that we should fork the code base
- freedom to hack --- getting our developers to "change vision"
- bugs in our GENI base are OK
- instability is OK
- a fork is a temporary thing; in one or two years, they will be remerged
- Jay: "I don't care that much if the Emulab base has quality degradation."
- Eric: is there a middle ground? Can we componentize the code,
and then fork only the parts that are needed?
- Rob: this would be in conflict with the !ProtoGENI "develop fast" constraint
- Rob: reasons to fork
- assign_wrapper: must change, with systemic effects; touching lots of bits
- Emulab and !ProtoGENI need different terminology; different user interfaces and other front ends
- Mike: forking is "the death of Emulab"
- fork represents that fact that we're moving on
- we just won't have time to maintain both
- Jay: what about the workbench?
- must run on regular Emulab, where we already have users
- so we must keep Emulab alive
- Final comments
- Jay: our !ProtoGENI work can be "successful" even if GENI fails
- Jay: generalizing Emulab is important in its own right
- Leigh: OK to table this for now
- eventually we will get to a point where fork versus non-fork matters
- Jay: free your mind to rearchitecting Emulab!
- Resolution: table the discussion for now
GENI Abstractions
- Rob: the stuff in the !ProtoGENI docs may not look a lot like what
we have now...
- ...but indeed, Rob believes that we can map a lot of the !ProtoGENI API onto our existing abstractions and implementation
- Leigh: not worried about implementing the basic !ProtoGENI API;
he's worried more about other issues
- virtualization
- with vservers
- security model: model is so poorly specified
- Jay: lead or follow in this area? Following is more work-efficient, but leading might be faster in the sense that we get what we like
- delegation is an intrinsic part of the model
- infrastructure required for handling keys
- Rob: implement something simpler than the full GENI
specification, because the specification may never be
fully fleshed out
- we implement a similar API
- come up with our own ways of doing certain things, e.g., signing
- we store the bags of bits for users
- virtualization
- Jay: what about collaborating with outside folks about security
stuff?
- Steve Schwab wants to collaborate with us
- Handling virtualization
- Jay: what about the shared node stuff?
- Leigh: we don't really have the concept of a shared node at
all in our core system
- we cannot share a node between projects
- Rob: the model that we already have for the PlanetLab nodes
is what we can go with
- component == physical node
- slivers == resources
- as far as the database goes, we're not that far from what we need
- Question: do we want !ProtoGENI to work on PlanetLab nodes?
- "break it and then fix it"
Node Manager
- David was going to look at the node manager stuff, but was on vacation and didn't have time
- Initial thoughts: do not take the schema wholesale; take some of their schema
- Otherwise, we can take a lot of stuff from the node manager
Action Items
- Leigh [carryover from previous meetings]: look at how we are currently unfaithful to XMLRPC or SOAP standards
- Rob: work more on software list
- Mike and Rob: list the features we don't have now to set up a
GENI experiment? "List of things that must be fixed."
- e.g., shared nodes other than PlanetLab nodes
- e.g., we don't have tunneling to PlanetLab nodes
-- Main.EricEide - 07 Nov 2007