6.896 Summary and Speculations

Hari Balakrishnan

EECS & LCSMIT

December 7,1998

Goals

• Gain exposure to classic and recent papers in networking Focus on protocols, architecture, applications Not on link technologies or hardware

• Learn how to do networking research; develop the ability to pick problems and critique research

• Conduct networking research via term project (and homework questions)

Results

• Studied about 50-odd papers across a wide-range of topics, in varying levels of detail

• Focused on key contributions and ideas, not details; focused on methodology and key results

• 17 different term projects An impressive array of topics and excellent

progress so far• Homework questions turned out being semi-

research projects too!

What Did We Learn in 6.896?

• Philosophy and Architecture• Methodology and Techniques• Congestion Control and Management

Data transport and end-to-end methods Queue management

• Routing Unicast Multicast Fast router design

• Mobility and Wireless

What Did We Learn in 6.896? (cont.)

• Applications HTTP; interactions between HTTP & TCP Adaptive multicast applications

• Re-thinking the Service Model Integrated and differentiated services Utilities, incentive structures, priorities

• Reliable Multicast• Programmable Networks• Middleware (Web Caching & DNS)• Network Security (IPSec)• Experimental Performance Analysis

Philosophy and Architecture

• Layering• Soft-state; fate-sharing• End-to-end argument• “Keep the interior simple”• ALF• IETF approach to standardization

“Rough consensus and working code”

Research Methodology

• Theoretical analysis (e.g., Bolot delay paper)• Design (e.g, using simulation)• Implementation• Experimental analysis and evaluation• Deployment issues

End-to-end Congestion Management

• Stability of AIMD schemes• Rate vs. window• DECBit scheme• TCP

Queue Management

• Scheduling vs. buffer management• Fair queueing• RED• Recommendations to encourage end-to-end

mechanisms

Unicast Routing

• Distance-vector vs. link-state• Intra-domain protocols (RIP, OSPF)• Inter-domain protocols (EGP, BGP)• Landmark hierarchy

Multicast Routing

• IGMP• DVMRP• CBT• PIM

High-Speed Routers

• BBN 50 Gbps router• High-speed forwarding techniques

Mobility

• Mobile IP• Routing• Weak authentication• Route optimization• Alternate approaches using…

Multicast Naming (say, DNS or better)

Wireless Transport

• Problems Errors Asymmetry Small windows

• Snoop protocol (handles errors)

Media-Access

• CSMA/CA and 802.11 “Listen-before-transmit” with exponential backoff

• MACA and MACAW RTS/CTS/DS/DATA/ACK

• Scalable channel access Scheduling Minimum-energy routing

HTTP

• Simple object download protocol• Multiple objects concurrent through

concurrent connections• Interacts badly with TCP congestion

mechanisms• Persistent connections and pipelining

(HTTP/1.1) Not the best solution!

Multicast Video

• IVS: Scalable feedback based on random sliding key scheme

• RLM: Layered video plus adaptive join/leave control mechanism

Integrated and Differentiated Services

• More than just best-effort• Reservation mechanisms (ISPN)• RSVP• Diff-serv models

Two-bit scheme with bandwidth broker Assured service

Utilities and Incentives

• Utility vs. rate (or delay) curves• Framework for arguing about right service

model• And arguing about admission control• Elastic vs. inelastic apps• Analysis for layered video and priority-drop

gateways

Reliable Multicast

• Key problem: message implosion• RMTP (designated receivers)• SRM (random damping + suppression and

ALF)• Digital fountain (no feedback, good FEC)• Replier-based scheme (router support for

local recovery)

Programmable Networks

• Eases deployment of new services• Active networks

General-purpose programmability Capsules Switchware

• Active services

Web Caching

• Ease load on servers and network, improve latency

• HTTP/1.1 support• Harvest• ICP• Summary Cache• Adaptive Web Caching• Cisco CacheDirector

Naming

• DNS• Future efforts

Active names Intentional names

Network Security

• IPSec• Intrusion detection • Public key infrastructure• Multicast key management

Performance Analysis

• Self-similarity and LRD• Synchronization of weakly-coupled

independent periodic processes• End-to-end Internet dynamics• SPAND: Shared and passive network

performance

A Sample of the Future

• Change caused by link technology, applications, scale, heterogeneity, pervasiveness

• Architecture & service model• Technological advances

Enormous diversity of wireless and wired links Satellite networks (poised to make a big splash!)

• Ubiquity Embedded devices Sensors & actuators Interactions with the world around us (“deeply

networked systems”)

Futures (cont.)

• Naming and Location Of objects, services, software, people, nodes, … E.g, Active naming; intentional naming Application-level routing

• Adaptation (protocols and apps) Congestion, load, errors, variability, mobility

• Self-configuration and self-organization “Spontaneous networks”

• Programmability Active nodes (in some form) almost certain

Futures (cont.)

• Novel services E.g., telephony, telepresence, graphics apps,

games• Security• Economic structure• Physics

Understanding the “physics” of networking and large-scale systems (like the Internet)

A global measurement and analysis infrastructure