140 likes | 298 Views
Deliverable. A meeting report that outlines our current thinking about Private PlanetLabs and Federation. Private PlanetLabs: Opportunities and Challenges Todo Pound out an outline (this afternoon) Distribute paragraph generation (tonight & tomorrow). Framework Questions. What is PlanetLab?
E N D
Deliverable • A meeting report that outlines our current thinking about Private PlanetLabs and Federation. Private PlanetLabs: Opportunities and Challenges • Todo • Pound out an outline (this afternoon) • Distribute paragraph generation (tonight & tomorrow)
Framework Questions • What is PlanetLab? • Operational system, software package (platform), arch, community model • What role PLC? • Ops, core development, specs, integration • Community • Innovating on top of PL • Contributing to PL itself • enablers: stable interfaces, myPLC, documentation) • Goals • Scale (parallelize/amplify PLC) • Enable community
Private PlanetLab Enable customization Localization Funding Policies Isolation Safe testing Solve local problems Distr. Virtualization Deploy services for local users Federate Scale ops Private regional + fed Need more resources (cycles + net effect) Join community (peers) Buy from providers Access to services Users, not just developers Altruism Why?
Risks • Divergent platform (harder to code to) • Divergent policies (prohibits certain slices) • Fragmented from user’s perspective (harder to secure needed resources)
Strategy to Mitigate Risks • Keeping common architecture makes federation is technically feasible • Keeping common code base lessens code diversity risk • Keeping common ops base lessens fragmentation risk (inter-NOC) • Keeping common policy lessens fragmentation risk (shared AUP) • Keep incentives in mind (market-based)
Software Development Issues • Types • Bug fixes • Generalizations (enable new capabilities) • Specializations/Enhancements (new capabilities) • Model • Patches / developer access (BSD model) • Limited by build/QA process • Borrows from Linux model • “Linus” decides (colored by operational requirements) • Intermediate testers/integrators (private planetlabs) • Well-known long-term architectural vision • Borrows from Amazon: Continuous release (node groups) • Sync’ing (upgrading) private PlanetLabs
Operational Support Issues • Training • Mgmt people (Best Practices) • User Education (and help desk) • Inter-NOC coordination • 3rd Party Management Services • Stork, CoBlitz, CoMon… • Sync’ing with PLC w/out disruption
Technical Challenges • PlanetLab/Cluster integration • Slice gets VM from the cluster • Slice gets a substantial chunk of the cluster • Slice a wireless access point • Slice a “gateway” to the Internet (e.g., BGP) • Feed “live” PL measurements into Emulab • Performance • Hardware-assist • Kernel mods
Technical Challenges (cont) • Scheduling to support low-jitter services • Generalize RSpec/NM • Alternative VM types (e.g., Xen) • Richer node types (e.g., ns -> XML) • Service composition • Dependencies across federated PLs
Friday’s Agenda • Source code base (Mark) • Clusters (Marc) • Lists • 6-12 month todo list (PLC and elsewhere) • Ops • Code • Arch/Documents • Wish list (longer term)
Discussion Points • Motivation/Rationale • Process-related issues • Software Development • Operational Support • Federation • Technical Challenges
More details on “Why” slide • Grow public PlanetLab • Scale operations beyond a single PLC • Make more aggregate resources available • Internationalization • Accommodate different policies • Can get funding from local sources • Manage distributed machines • Common point-of-pain w/ minimal “public PL” impact • Pair-wise peering will naturally follow • new communities spring up • Specializing/Extending • Researchers want a playground • Deploying useful services in new environs
Federation Issues • Refine the architecture/interfaces • Peering agreements • Right resource control/allocation knobs • AUP • Transitioning to independent MAs/SAs • Common (seamless) user experience • Internationalization • Membership agreement • Differing AUPs