Arguments for Separating Edges from Transit Core by Dan Jen, Michael Meisel, He Yan, Dan Massey, Lan Wang, Beichuan Zhang, Lixia Zhang presented by Dan Jen University of California, Los Angeles 1
Routing Scalability is a Hot Topic 2
Will Routers Be Able to Keep Up? • No consensus – Yes! – No! • Correct Answer: – Maybe... but what if they can't? 3
BSTS. IRTF RRG. • Routing Research Group – “One of the key topics facing the Internet today is the issue of Interdomain routing scalability - the need for a new routing architecture. Currently, a sub-group of the IRTF RRG has been created to define and investigate what the possible requirements for such an architecture might be and to also explore possible methods.” • Basically, come up with a proposal that helps scale the Internet for the foreseeable future. 4
Many Scalability Proposals, No Verdict yet APT TRRP Six/One Router Multipath Transport LISP ILNP Shim6 5
Many Scalability Proposals, No Verdict yet APT TRRP Six/One Router Multipath Transport LISP ILNP Shim6 Can We Help with the Debate? 6
Our Contributions Separation Elimination APT Shim6 LISP Multipath Transport TRRP ILNP Six/One Router 7
Our Contributions Separation Elimination APT Shim6 LISP Multipath Transport TRRP ILNP Six/One Router 8
Separation > Our Contributions Elimination Shim6 APT Multipath Transport LISP TRRP ILNP Six/One Router 9
Observation: Two Camps • All proposals use aggregation as the mechanism for routing scalability. • Ability to aggregate(and thus scale) comes in one of two ways: – Separate non-aggregatable prefixes from global routing scope. – Eliminate the use of non-aggregatable addresses altogether. 10
Separation • Edge networks moved outside the scope of global routing. Only transit network prefixes are globally routable. • Edge networks are mapped to routable, aggregatable provider addresses. – Each Separation proposal has a mapping system Example: Map & Encap 11
Dst Src Site2 Site2 Site1 Site1 Edge Space Edge Space Transit Core Transit Core P2 P1 ISP1 ISP1 ISP2 ISP2 12 ISP3 ISP3
Dst Dst Src Site2 Site2 Site1 Site1 Edge Space Edge Space Transit Core Transit Core P2 P1 ISP1 ISP1 ISP2 ISP2 13 ISP3 ISP3
Dst Src Site2 Site2 Site1 Site1 Dst Edge Space Edge Space Transit Core Transit Core P2 P1 ISP1 ISP1 ISP2 ISP2 14 ISP3 ISP3
Dst Src Site2 Site2 Site1 Site1 Edge Space Edge Space Transit Core Transit Core Dst P2 P1 ISP1 ISP1 ISP2 ISP2 15 ISP3 ISP3
Dst Src Site2 Site2 Site1 Site1 Edge Space Edge Space Transit Core Transit Core P2 Dst P2 P1 ISP1 ISP1 ISP2 ISP2 16 ISP3 ISP3
Dst Src Site2 Site2 Site1 Site1 Edge Space Edge Space Transit Core Transit Core P2 P1 P2 Dst ISP1 ISP1 ISP2 ISP2 17 ISP3 ISP3
Dst Src Site2 Site2 Site1 Site1 Edge Space Edge Space Transit Core Transit Core P2 Dst P2 P1 ISP1 ISP1 ISP2 ISP2 18 ISP3 ISP3
Dst Src Site2 Site2 Site1 Site1 Edge Space Edge Space Transit Core Transit Core Dst P2 P1 ISP1 ISP1 ISP2 ISP2 19 ISP3 ISP3
Dst Src Site2 Site2 Site1 Site1 Dst Edge Space Edge Space Transit Core Transit Core P2 P1 ISP1 ISP1 ISP2 ISP2 20 ISP3 ISP3
Dst Dst Src Site2 Site2 Site1 Site1 Edge Space Edge Space Transit Core Transit Core P2 P1 ISP1 ISP1 ISP2 ISP2 21 ISP3 ISP3
Elimination • Keeps all players in routable scope, but eliminates the use of unaggregatable prefixes • Hosts are upgraded with the ability to utilize multiple IP addresses. Example: Multipath Transport 22
1.2.1.1 3.4.1.1 Multihomed Multihomed host1 Edge Site1 Edge Site1 1.2.1/24 3.4.1/24 ISP2 ISP2 ISP1 ISP1 3.4/16 1.2/16 1.2.2/24 3.4.2/24 Multihomed Multihomed host2 Edge Site2 Edge Site2 23 1.2.2.1 3.4.2.1
1.2.1.1 3.4.1.1 Multihomed Multihomed host1 Edge Site1 Edge Site1 1.2.1/24 3.4.1/24 ISP2 ISP2 ISP1 ISP1 3.4/16 1.2/16 1.2.2/24 3.4.2/24 Multihomed Multihomed host2 Edge Site2 Edge Site2 24 1.2.2.1 3.4.2.1
1.2.1.1 3.4.1.1 Multihomed Multihomed host1 Edge Site1 Edge Site1 1.2.1/24 3.4.1/24 ISP2 ISP2 ISP1 ISP1 3.4/16 1.2/16 1.2.2/24 3.4.2/24 Multihomed Multihomed host2 Edge Site2 Edge Site2 25 1.2.2.1 3.4.2.1
Our Claim: Separation is Better • Who deploys scalability solution? – Separation solutions deployed by networks – Elimination solutions deployed by hosts • Who benefits from scalability? – Networks only 26
Elimination: Misalignment of Cost and Benefit • Must Fight Inertia – Hosts don't directly benefit from routing scalability – Other incentives unrelated to scalability offered by elimination proposals to convince hosts to upgrade. – Deployment speed depends upon this incentive's desirability. • No guarantee that enough hosts will upgrade before it's too late. 27
Separation Ties Demand with Deployment • Under separation, affected parties are the ones with the power to fix their problems. • Networks that need scalability can deploy the new solution and benefit immediately – APT, basically an interdomain Viaggre (and more!) – What's Viaggre? See next presentation. 28
Other Reasons for Choosing Separation • Easier Rollout of New Protocols – Edge Networks and Transit Networks can deploy new protocols independently of one another. • DDoS Mitigation – Mapping System is a natural mounting point for DDoS defense • Explicit Ingress Traffic Engineering – Can be placed in mapping entries. – No more need for prefix splitting and BGP trickery * Details Are In The Paper * 29
(Separation > Elimination) != (Elimination Schemes are Useless) • Use separation for scalability • But retain non-scalability benefits offered by elimination schemes. – Example: Multipath transport • Often, these benefits can be added to separation schemes • Again, details are in paper. 30
1.2.1.1 3.4.1.1 Multihomed Multihomed host1 Edge Site Edge Site 1.2.1/24 3.4.1/24 ISP2 ISP2 ISP1 ISP1 3.4/16 1.2/16 1.2.2/24 3.4.2/24 Multihomed Multihomed host2 Edge Site Edge Site 31 1.2.2.1 3.4.2.1
In Conclusion... 32
Before our work, the Routing Research Group argued endlessly about all of the scalability solutions on the table. 33
Now, we hope the Routing Research Group can stop arguing about Elimination solutions to scalability, and argue endlessly about other things. 34
Questions? Comments? jenster@cs.ucla.edu 35
Recommend
More recommend
