Bi-Level Active Networks

	Jon Crowcroft
	Hermann De Meer
	Ken Carlberg

Background: IP

	IP architecture built on Layered Model
	Some current efforts bend/break the model
		Congestion avoidance: RED
		Division of resources: CBQ, WFQ
		QoS: diff-serv, int-serv

Background: Active Networks

	Add functionality/services “on the fly”
	Initially, Active Router (AR)
		‘Reprogram’ code in data packets
	Active Server (AS)
		Dynamically loaded software modules
		Used as proxies for additional services offered by the network
			Transcoding
			Dynamic VPNs

Active Net Issues

	Security
		Need to prevent adverse functions
		Approaches include strict semantics (SecuNet), polices (Android/FAIN), and ‘sandboxes’ (execution environments)
	Performance
		‘Slow path’ for option-filled data packets in routers
		Lack of inter-layer information leads to…
			Redundant functions between layers (e.g., Application layer routing/discovery)
			Sub-optimal performance of end-to-end flows

Objectives

	Retain the layered model
		Break it for ‘active’ flows
	Focus solution on integrated operation between AR and AS

Bi-Level Architecture

	Network consisting of AR and AS
		Similar to ALAN architecture
		AS and AR communicate/compliment each other
	Language between AR and AS
		Articulate type, form, target
		Expansion of Secunet language(?) and/or XML-programable router project(??)
	Policies

Architecture Example

Bi-Level Usage Example

	Active Flow through AS-A and AS-B
	AS-A determines that flow is subject ‘active’ service
	AS-A triggers AR to send constrained flood
		Determine new paths with different metrics
		Constrain advertisement to shaded region using ‘active’ routing protocol
	Note: Example represents
	gradual active service

Another Example

	NATs and Firewalls
	Nat’s operate like a primative network AN device
	Firewalls act like an application layer (possibly)
	Need to coordinate (e.g. the tweaking/open door protocol) between the two

Reflection on Action,
Action on Reflection

	Reflection is a neat tool for distributed systems middleware
	C.f. work at lancaster and irvine on reflection for mobile and multimedia
	Seems ideal for Bi-Level architecture too!

Summary/Conclusions

	Good reasons for 2 levels of overlay
	Can achieve 3 or more just by recursing, so no need for n-layer model, unlike ISO
	Need to work on the meta-object protocol in the real sense of the word protocol so that two layers can comprehend each others internal behaviours.