Pathlet routing_Demo2
- 1. PATHLET ROUTING Amit Srivastava Gauri Pulekar
- 2. Introducing, ‘Pathlet Routing’ ACM SIGCOMM Computer Communication Review 2009 Proposed by research team from UC Berkeley and UI Urbana Champaign Addresses two main problems with BGP Scalability Multipath Routing Allows protocol emulation, it can mimic BGP
- 3. BGP : It runs the internet! Current de-facto inter AS routing protocol BGP uses destination based forwarding i.e. IP address in the header Uses TCP connection to peers for sending information Only shares a portion of the routing table No periodic update to routing information Routers in the same AS are internal peers, others are external peers
- 4. BGP : Messages KEEPALIVE message sent for checking connectivity NOTIFICATION are for failure or before closing connections UPDATE messages carry routing information Routes are stored in Routing Information Bases (RIB): Adj-RIB-In contains information from inbound UPDATE messages, mostly unprocessed information Loc-RIB contains local routing formation by applying local policies on the data in Adj-RIB-In Adj-RIB-Out contains the routing info the router has selected to share with its peers via UPDATE
- 5. Building Blocks Vnodes: Virtual node For a given router, it can have n vnodes for n interfaces This allows traffic shaping for traffic coming from an interface Routers learn about vnodes in the AS by exchanging information with peers Pathlets: A sequence of vnodes leading to a destination identified by a IP prefix Forwarding Identifier – FID (f) Vnode + Pathlet = Virtual topology
- 6. Let’s understand.. Constructing pathlets between peer routers A B C 2
- 7. Let’s understand.. Two pathlets used to send data from A to C 3 2 A B C 2, 3 3
- 8. Let’s understand.. Three autonomous systems 3 2 A B C 4 4 3
- 9. Let’s understand.. Three autonomous systems A B C y v r q s
- 10. Let’s understand.. Ingress and egress points can be chosen from many available options A B C x v r q s
- 11. Let’s understand.. Multiple pathlets to suit local transit policies A B C x v r q s
- 12. Pathlet Routing - Working Pathlet originates at a router R1 (with vnode v1) The sending vnode refers to its table for possible pathlets to reach destination Creates a list of vnodes and calls it a pathlet ( and can advertise it) puts the pathlet in the header and send the packet out to a peer Peer removes its own vnode in the packet header and sends along to the next vnode This path selection is dependent on local policies of the AS
- 13. Route Selection Using shortest path algorithm Path for packet forwarding based on result of shortest path algorithm run on a graph created by using routing information at vnodes
- 14. Pathlet Dissemination Path vector algorithm (just like BGP) Announces pathlet’s FID and sequence of vnodes Scalability: Propagate arbitrary subset of known pathlets Path vector used for sending routing info not routing data
- 15. Components Of Pathlet Routing
- 16. Components Of Pathlet Routing Controller: Processes the routing information Vnode Manager: Manages all vnodes(tables) on a single device Disseminator: Sends out routing information to peers Router : Router running at least 1 pathlet routing instance Vnode: A small routing table Pathlet: A sequence of vnodes
- 17. Implementation: Router The router can have max 5 interfaces It can connect to other routers using TCP The topology is created using a config file There are 2 kinds of messages Data messages with dummy data Routing messages contain routing information in the following format {AS-id, RouterId, VnodeId} and {PathletId (FID), IP Prefix} Routers collect stats like size of table in Vnodes and total control messages sent and received Allow commands on each router
- 18. Implementation: Vnode & Vnode Manager Vnode Manager is a container for Vnodes All VNodes have vnode-id, router-id, AS-id to uniquely identify it Controller logic is still being decided but it only decides on installing paths from peers on Vnodes The shortest path algorithm is implemented on the Vnode The shortest path algorithm implemented separately but buggy
- 19. Implementation: Deciding Topology Using BGP RIB and Update messages to create update messages A simple 3 and 6 node topology being used to achieve basic goal of routers exchanging data Currently tables are static
- 20. Scope Implement the software router in Python using Twisted and NetworkX. Allow exchange of messages on routers arranged in real-world like scenario Save the logs for comparison with data from RouteViews Simulate loss of links
- 21. Timeline Activity Completed Time Find internet-based topology data or a random graph which can be scaled from 10 to 300 nodes 20 October Learn and implement the shortest path and Path Vector algorithms 20 October Learn basic Erlang. We are using Python with NetworkX and Twisted 27 October Explore possible packet structure for Pathlet routing 27 October Finalize the details to be implemented or BGP and Pathlet routing 3 November Activities completed so far
- 22. Timeline Ongoing activities and future plan Ongoing Activities Time Run the algorithms for Pathlet and BGP on a large 100+ node graph. And verify the result 10 November Simulate of a 3 node graph running BGP and Pathlets capable of exchanging information. Created a graph of 6 nodes to implement Pathlet only not BGP 17 November Verify the above system for discrepancies 24 November Add provision for emulating Local policies 1 December for implementing Local Transit policies Run comparison studies on the system 8 December
- 23. References http://dl.acm.org/citation.cfm?id=1592583 https://www.ietf.org/rfc/rfc1771.txt http://jad-reads.blogspot.com/2009/04/thoughts-on-pathlet-routing. html ftp://espectrocrom.com/pub/manuales/bgp/cisco%20- %20bgp%20- %20internet%20routing%20architectures%202ed%20(press%201- 57870-233-x).pdf http://web.engr.illinois.edu/~pbg/pathlets/
- 24. Questions
- 25. Thank You