1. Vytautas Valancius, Nick Feamster, Akihiro Nakao, and Jennifer Rexford

2.  Cloud computing is on the rise  Provides computing resources and storage in cloud data centers  Hosting on the steroids for Internet services 2

3. 3 Cloud Data Center Data Center Router Interactive Service Bulk transfer Internet Routing updates Packets ISP1 ISP2  Hosted services have different requirements  Too slow for interactive service, or  Too costly for bulk transfer!

4.  Multiple upstream ISPs  Amazon EC2 has at least 58 routing peers in Virginia data center  Data center router picks one route to a destination for all hosted services  Packets from all hosted applications use the same path 4

5.  Obtain connectivity to upstream ISPs  Physical connectivity  Contracts and routing sessions  Obtain the Internet numbered resources from authorities  Expensive and time-consuming! 5

6. 66 Cloud Data Center Interactive Service Bulk transfer Internet ISP1 ISP2 Virtual Router B Virtual Router A Transit Portal Routes Packets Full Internet route control to hosted cloud services!

7.  Motivation and Overview  Connecting to the Transit Portal  Advanced Transit Portal Applications  Scaling the Transit Portal  Future Work & Summary 7

8.  Separate Internet router for each service  Virtual or physical routers  Links between service router and TP  Each link emulates connection to upstream ISP  Routing sessions to upstream ISPs  TP exposes standard BGP route control interface 8

9. Transit Portal Virtual BGP Router 9  Cloud client with two upstream ISPs  ISP 1 is preferred  ISP 1 exhibits excessive jitter  Cloud client reroutes through ISP 2 ISP 1 ISP 2 Interactive Cloud Service BGP Sessions Traffic

10.  Server with custom routing software  4GB RAM, 2x2.66GHz Xeon cores  Three active sites with upstream ISPs  Atlanta, Madison, and Princeton  A number of active experiments  BGP poisoning (University of Washington)  IP Anycast (Princeton University)  Advanced Networking class (Georgia Tech) 10

11.  Internet services require fast name resolution  IP anycast for name resolution  DNS servers with the same IP address  IP address announced to ISPs in multiple locations  Internet routing converges to the closest server  Available only to large organizations 11

12. ISP1 ISP2 ISP3 ISP4 Transit Portal AsiaNorth America Anycast Routes 12 Name Service  TP allows hosted applications use IP anycast

13.  Internet services in geographically diverse data centers  Operators migrate Internet user’s connections  Two conventional methods:  DNS name re-mapping ▪ Slow  Virtual machine migration with local re-routing ▪ Requires globally routed network 13

14. ISP1 ISP2 ISP3 ISP4 Transit Portal AsiaNorth America Tunneled Sessions 14 Active Game Service Internet

15.  Scale to dozens of sessions to ISPs and hundreds of sessions to hosted services  At the same time:  Present each client with sessions that have an appearance of direct connectivity to an ISP  Prevented clients from abusing Internet routing protocols 15

16.  Conventional BGP router:  Receives routing updates from peers  Propagates routing update about one path only  Selects one path to forward packets  Scalable but not transparent or flexible 16 ISP1 ISP2 BGP Router Updates Client BGP Router Packets

17. Bulk Transfer Routing Process  Store and propagate all BGP routes from ISPs  Separate routing tables  Reduce memory consumption  Single routing process - shared data structures  Reduce memory use from 90MB/ISP to 60MB/ISP 17 ISP1 ISP2 Virtual Router Routing Table 1 Routing Table 2 Interactive Service

18. Bulk Transfer Routing Process  Hundreds of routing sessions to clients  High CPU load  Schedule and send routing updates in bundles  Reduces CPU from 18% to 6% for 500 client sessions 18 ISP1 ISP2 Virtual Router Routing Table 1 Routing Table 2 Interactive Service

19. Forwarding Table  Connecting clients  Tunneling and VLANs  Curbing memory usage  Separate virtual routing tables with default to upstream  50MB/ISP -> ~0.1MB/ISP memory use in forwarding table 19 ISP1 ISP2 Virtual BGP Router Forwarding Table 1 Forwardng Table 2 Bulk Transfer Interactive Service

20.  Future work:  More deployment sites  Making TP accessible for network research test-beds (e.g., GENI, CoreLab)  Faster forwarding (NetFPGA, OpenFlow)  Lightweight interface to route control 20

21.  Limited routing control for hosted services  Transit Portal gives wide-area route control  Advanced applications with many TPs  Open source implementation  Scales to hundreds of client sessions  The deployment is real  Can be used today for research and education  More information http://valas.gtnoise.net/tphttp://valas.gtnoise.net/tp 21 Questions?

22.  Used in a “Next-Generation Internet” Course at Georgia Tech in Spring 2010  Students set up virtual networks and connect directly to TP via OpenVPN  Live feed of BGP routes  Routable IP addresses for in class topology inference and performance measurements 22

23. 23  Three active sites  Princeton, NJ  Atlanta, GA  Madison, WI  Internet numbered resources from ARIN  AS 47065  IPv4 prefix: 168.62.16.0/21