Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

20100120 Pvm Cherchez La Federation


Published on

  • Be the first to comment

  • Be the first to like this

20100120 Pvm Cherchez La Federation

  1. 1. Cherchez La Federation or How d we motivate federation? H do i f d i ? Paul Mockapetris (, Serge Fdida (serge fdida@lip6 fr) ( Panayotis Antoniadis (
  2. 2. A simple definition for Federation  Federation occurs when two or more organizations agree to each allocate some of their resources to implement a common service. Org Org A B 2
  3. 3. Federation Examples I t Internet t – Share connectivity and multiple levels of protocol  BGP – Share routing information  DNS – Distribute configuration g  Carter phone Decision – Create interface for communication devices 3
  4. 4. Qui Bono?  Th The reasons f federation revolve for f d ti l around mutual benefit, but in practice are quite different f different t it diff t for diff t types of f organizations.  Corollary:a particular federation or federation technology can be attractive or unattractive to different organizations organizations. 4
  5. 5. Governments G l Goals: – Maximize overall economy – Maximize overall user satisfaction – Minimize risk  Methods: – Regulate monopolies; encourage competitiveness; smaller and more numerous markets – Encourage best of breed solution finding 5
  6. 6. Enterprises G l Goals – Persistent competitive advantage – Leverage strategic advantages to new markets – Reduce risk  Methods – Own monopolies or near monopolies – Commodity i C di inputs; proprietary outputs i 6
  7. 7. Users G l Goals – Best technology – Best Price – Ease of use / Least investment in learning g  Methods – Purchasing h i P h i choices – Influence on Government 7
  8. 8. Researchers G l Goals – Innovation – Freedom to explore alternatives  Methods – Joint efforts to aggregate scarce resources – Embrace di E b disruptive risk ti i k 8
  9. 9. Federating Clouds (Researchers) M k Makes eminent sense for researchers i t f h – Scale of all academic resources less than that of production systems by orders of magnitude • E.g. PlanetLab, OneLab • Use individual’s phones as distributed sensor net, or intermittently connected testbed 9
  10. 10. Federating Clouds (Enterprise Providers)  Littl Little advantage t d f Cl d providers d t today for Cloud id – Makes their output a commodity – Scale isn’t an issue – Coverage isn’t an issue (y ) g (yet?)  Perhaps federate lower levels? – (federate (f d t commodities, not discriminators) diti t di i i t ) 10
  11. 11. Federating Clouds (Enterprise users) C Can we use f d ti t reduce risk? federation to d i k? – Avoid cloud lock in – Handle demand surge  Connect different cloud apps – Inevitable once separate cloud-based apps get important enough? 11
  12. 12. Federating Clouds (Governments) G Governments t as big enterprises bi t i – Lots of resources to share & save – Lots of data to make available, internally and externally – Load moves around (e.g. IRS in April?)  Governments as regulators – Policy? 12
  13. 13. Conclusions L t Lots of forces work against federation ff k i t f d ti  Will happen first in pp – Research world – Governments  We want to make consolidation happen sooner b t th ’ a growing b but there’s i base of f legacy inertia. 13
  14. 14. Questions H How t shape academic efforts so that to h d i ff t th t they will be relevant to the other worlds? – E.g. Agenda and participation in NSF workshop  Isit the same as networks before the Internet? 14
  15. 15. Today vs.1983? y (ARPAnet extinction; start of the Internet age)  Si il iti Similarities – Lots of production proprietary inertia  Differences – Ease of experimentation with multiple open standard OSes, etc. – Cheap hardware – End of the hourglass model 15
  16. 16. End of the hourglass?  IP is th center of the Network just as the earth i the t f th N t kj t th th is the center of the universe.  It’s ’still the most important f frame of reference, f f but not the only one.  E.g. – Ethernet may be the way to port servers in datacenters – Custom content based networks need not be IP based. based 16
  17. 17. The Gravitational model W We have large adoption ( h l d ti (mass) of certain ) f t i standard protocols, uses and interfaces. – IP, IP users, movies, web, games, … i b  There’s a huge attraction to finding a better path b t th between any t two of these (so long as you f th ( l don’t fall into another gravitational well)  This leads to multiple paths between the biggest masses, or a polymorphic net. 17