Capsule-based Active Networks: What have we learned?

13
Capsule-based Active Networks: What have we learned? David Wetherall ([email protected]) University of Washington

description

Capsule-based Active Networks: What have we learned?. David Wetherall ([email protected]) University of Washington. State of Active Networks. AS1. ACTIVE IP. Plugins. Softnet. ANTS. ??. M Ø. PLAN. The Active Internet. PAN. Vision. 96. 97. 98. 99. 00. …. 0x. - PowerPoint PPT Presentation

Transcript of Capsule-based Active Networks: What have we learned?

Page 1: Capsule-based Active Networks: What have we learned?

Capsule-based Active Networks: What have we learned?

David Wetherall ([email protected])University of Washington

Page 2: Capsule-based Active Networks: What have we learned?

djw // OpenSig’99 // 10/99 2

State of Active Networks

So when canI use it?

Vision96 999897 00

The ActiveInternet

ANTSPLAN

ACTIVE IPMØ PAN

PluginsAS1

Are we there yet?

??Softnet

0x…

Page 3: Capsule-based Active Networks: What have we learned?

djw // OpenSig’99 // 10/99 3

Context: Styles of Extensibility

System User

Middleware transcoding proxies (AS1)

mobile agents(Active Names)

Packet congestion gateway (Plugins)

routing w/ nack suppression

(PLAN / ANTS)

Getting beyond “One size fits all”

Page 4: Capsule-based Active Networks: What have we learned?

djw // OpenSig’99 // 10/99 4

This is ANTS

• Aim: Each user controls the handling of their own packets in the network

• How well does this work?

User 1

User 2

capsules

activenodes

Page 5: Capsule-based Active Networks: What have we learned?

djw // OpenSig’99 // 10/99 5

ICMP as a Minimal System

• ICMP– Runs at all IP routers, can be used by all users

• Programs– Traceroute, path MTU discovery

flexibilityperformancesecurity ICMP ANTS?

Page 6: Capsule-based Active Networks: What have we learned?

djw // OpenSig’99 // 10/99 6

ANTS Tradeoffs

• Mobile code for flexibility• Simple forwarding model for performance /

security– Depend on caching; target experimentation

data IP

#7

#23

#7

RU

NTIM

E

Page 7: Capsule-based Active Networks: What have we learned?

djw // OpenSig’99 // 10/99 7

Accepting Heterogeneity

• Can’t make forwarding simpler, so processing ultimately depends on node’s abilities /position– E.g. 1GHz, 1Gbps, 1000b, 100% 1000 cycles– Or if 100Mbps, 10% 100,000 cycles– Wide variation heterogeneity is fundamental

• Build into programming model– Can query node to determine what’s

appropriate– Non-active nodes (IP routers) at bottom

Page 8: Capsule-based Active Networks: What have we learned?

djw // OpenSig’99 // 10/99 8

A Minor Matter of Security

• Vision: unprivileged users have unrestricted access

• Requires Protection– My program can’t corrupt your program We know how to do this

• Requires Resource Management– My program can’t starve your program The harder problem

Page 9: Capsule-based Active Networks: What have we learned?

djw // OpenSig’99 // 10/99 9

Resource Management Tasks

Internet ANTS

Per PacketAt One Router

Yes, by design of IP

Yes, for a simple model

Per Packet Yes, by design of protocols

Woops … big hole here

Per User Need FQ or penalty boxes

Same as Internet

Page 10: Capsule-based Active Networks: What have we learned?

djw // OpenSig’99 // 10/99 10

The “Per Packet” Problem

• TTLs are a weak solution; not related to topology• Fairness mechanisms mitigate, but arguably not

enough• Restricted program forms may help …

Want toStop this!

Page 11: Capsule-based Active Networks: What have we learned?

djw // OpenSig’99 // 10/99 11

Programming with ANTS

• Examples– Multicast (single source and PIM)– Mobility– Reliable Multicast support– Path MTU discovery– Auctions– Web cache diversion routing– “TCP-SYN” filtering

Page 12: Capsule-based Active Networks: What have we learned?

djw // OpenSig’99 // 10/99 12

Lessons / Discoveries

• How to program mixed networks: A complication• Code as flexible “glue”: Many variations• A clean way to query the network: Heterogeneity• The set of unchangeable things: Shared

assumptions

More Application Experience is Needed!

Page 13: Capsule-based Active Networks: What have we learned?

djw // OpenSig’99 // 10/99 13

More Details

• See paper in SOSP’99:

• Active Network Vision and Reality: Lessons from a Capsule-based System, D. Wetherall.