Active Networking and End-to-End Argument

Samrat Bhattacharjee

Kenneth L. Calvert

Ellen W. Zegura

Objective

End-to-End Argument Active Networking – Extension of E2E argument End System approach Vs Combined System

approach Performance Model to quantify the above

approaches

E2E and Active Networking

What is E2E? Is Active Networking a natural consequence of

E2E? E2E and placement of functionality

All applications might not use the serviceTrade-off between performance and costCombine network and application information to

optimize performance

Model for Service Location

Analyze performance under two design options – Design X : Service implementation exclusive in the

end-systemsDesign C : Service achieved through combination of

implementation at the end-system and in the network

Network is treated monolithically Network support is boolean

Model for Service Location

Parameters of performance modelExclusively End-system (Design X)

• Tx – Expected Performance

Combined End-system and Network (Design C)• Tc – Expected performance

• Pn – Probability that the network support accomplishes the service

• Te – Expected performance, end-system version

• Tn – Expected performance, network version

Tc = (1-Pn) Te + Pn Tn

Reliable Data Transfer

Performance Metric : Expected Transfer Time Design X :

tx – time to request, receive and check the integrityp – probability of error in each transmissionTx – expected transfer time

Tx =

i=1

P(i transmissions) * i * tx

= tx / (1-p)

Reliable Data Transfer ...

Design C tc – time to request, receive and check the integrityp – probability of error in each transmissionq – probability that the network can correct the error

Pn Tn = (1 – p + pq) tc Te = tc +

i=1

P(i transmissions) * i * tc

= tc (1+1 / (1-p+pq) ) Tc = (1 – p + pq) tc + p(1-q) * tc (1+1 / (1-p+pq) )

Reliable Multicast

Performance metric : Latency (no of hops) Design X

Buffering and Retransmission done only at the Receivers Request message is directed to a “nearby” Receiver through the

Loss node

Design C Buffering and Retransmission done by the network nodes

Reliable Multicast ...

Reliable Multicast ...

Tx = tR+ tL + tY + 2 tR’ + tE + tR

TN = 2(tR+ tL + 1)

TE = 2(tR+ tL + tL’ + tS)

Tc = 2pn (tR+ tL + 1) + 2 (1-pn) (tR+ tL + tL’ + tS)

assume tR = tR’ = tS and tL = tL’

Tx = 4tR+ tL + tY + tE

Tc = 4tR+ 4tL - 2pn (tR+ tL - 1)

If pn > (3tL- tE - tY ) / (2(tR+ tL - 1))

then Tc < Tx ( combined system approach is better)

Congestion Control

Application knows how to adapt Network knows where and when to adapt Flow packets contain advice about how to control

congestion and may be stored at the network node

Best Effort MPEG Delivery

Partial Packet Discard - discard packets on buffer overflow

Static Priority Discard - two level priority scheme Frame Level Discard – queue a datagram iff its

corresponding frame can be entirely queued Group of Picture Level Discard – if I-frame is

dropped, drop corresponding P & B frames.

Performance Analysis

Performance metric :

fraction of received data not discarded

Di,k – fraction of

discarded data

Ti,k – performance of

model i at a source rate of k Mbps

i = { P, S, F, G }

Performance Analysis

Performance metric : signal-to-noise ratio

Conclusion

Active networking is consistent with, and even suggested by the E2E argument

Active networks outperform the end-to-end solutions ?