Supporting Legacy Applications over i3

1
Supporting Legacy Applications over i3

• Jayanthkumar Kannan, Ayumu Kubota (VIF/KDDI),
  Karthik Lakshminarayanan, Ion Stoica and
• Klaus Wehrle (Univ. of Karlsruhe)

2
Motivation

• Proliferation of overlay networks, yet
• no common interface
• not many native applications
• not many deployments
• Chicken-and-Egg problem
• Requirements and methodology
• Unmodified legacy applications should work with
  overlay
• Two-step methodology
• Implement overlay using i3s flexible primitives
• Interface legacy applications to i3
• i3 chosen for its flexibility

3
Motivation

• Proliferation of overlay networks, yet
• no common interface
• not many native applications
• not many deployments
• Chicken-and-Egg problem
• Requirements and methodology
• Unmodified legacy applications should work with
  overlay
• Two-step methodology
• Implement overlay using i3s flexible primitives
• Interface legacy applications to i3
• i3 chosen for its flexibility

4
Overview
Insert middle-box apps
IP?i3 proxy
i3
i3?IP proxy
Internet
Transparent access across Firewall/NAT
Communicate with i3-unaware hosts via i3
mobility
5
Proxy Modes of operation

• Local proxy at both ends
• IP ?? IP communication over i3
• Local proxy at client and i3 ? IP proxy for IP
  server (say cnn.com)
• i3 ? IP proxy inserts triggers, performs flow
  setup on behalf on cnn.com
• Local proxy at server and IP ? i3 proxy for IP
  client
• foo.i3.6to4.jp mapped to foo.i3
• DNS server returns address of IP ? i3 proxy which
  performs flow setup

6
How the proxy works
foo.i3 ? ID-foo bar.i3 ? ID-bar .
. .
i3 client proxy
foo.i3
4. create a virtual host for foo.i3 with fake IP
address 10.1.2.3
Legacy apps
10.0.0.1
Legacy apps
Legacy apps
Linux/Windows
Server
Client
7
Benefits of our approach

• Local hijacking of flows
• Flexible, fine-grained decision of how flows
  should be routed
• Can be used for multi-homing, multi-path routing,
  etc.
• Addresses one of the big complaints of VPNs today
• Identifying hosts/services using IDs
• Name resolution layer is necessary anyway
• Stitch multiple address space (need to get around
  NATs)
• Security (large ID space)
• End-host control over routing
• Redirect through middleboxes (IDS, Spam
  detection, OverQos)

8
Resources

• URL i3.cs.berkeley.edu
• Source code
• Precompiled executables for Linux and Windows