Transcript presentation source

The Direction of Value Flow in
Connectionless Networks
Bob Briscoe
BT Research
19 Nov 1999
context
context
pricing
clearing
scenarios
summary
why think about this?
 theory of the value of communications:
• connection-oriented hegemony
[ITU96, Zull97, Clark96, MacKie-Var92]
 confusion is abroad in the world
• junk advertising, bundling content+network
 why assume multicast sender pays?
 alternative proposal seems obvious
• but wasn't when I started
19 Nov 1999
Direction of Value
3
context
context
pricing
clearing
scenarios
summary
context
 multiservice, multicast IP (real-time apps)
 some services usage-charged, whether
packet, flow or even longer term granularity
[Briscoeb99]
 electronic tariff announcement
[Rizzo99, Yemini98, Carle98]
 most general business model possible?
 buyer's market
 designed for bundling
• will cover in a future paper...
19 Nov 1999
Direction of Value
4
context
context
pricing
clearing
scenarios
summary
new model for what business?
end-system
applic’n
applic’n
present’n
present’n
session
e-commerce
transport
network
transport
e-commerce
network router
network
link
link e-commerce
physical
19 Nov 1999
session
physical
network
link
e-commercelink e-commerce
physical
Direction of Value
physical
5
context
context
pricing
clearing
scenarios
the value of place
value1 = f(i, place1 , time1 )
summary
i
value2 = f(i, place2 , time2 )
 xmt value /leg = (value of place-time)
i
i
 but competition  cost+ margin pricing
 who can know the value of place?
N: provider
Y: end-users
–  apportionment
 clearing grows comms market
19 Nov 1999
Direction of Value
6
context
pricing
pricing
clearing
scenarios
summary
end to end pricing
data flow
service
price
ISPs
B
customer
A
19 Nov 1999
end-to-end
pricing
Directionrole
of Value
C
8
context
pricing
pricing
clearing
scenarios
summary
end-to-end pricing role
 not necessarily a business; a role taken by:
– one ISP
– the customer
– a third party (+ bundling content?)
 openness to commercial innovation
 but…
...what is the best default apportionment?...
19 Nov 1999
Direction of Value
11
context
pricing
pricing
clearing
scenarios
summary
default value apportionment
 proposal:
– 'sender and receiver both charged'
– slight discrepancy in prices possible
– every leg of every flow is part charged twice
– model self-similar for inter-domain boundaries
19 Nov 1999
Direction of Value
12
context
pricing
pricing
clearing
scenarios
summary
split-edge pricing - per direction
Nd
Ne
Nd

Nb

Wbas
Wabr
Na


Wbar
Wabs
Nc
Nc
19 Nov 1999
 price in & out
Direction of Value

separately
each price
between boundary
and remote edge
price effects
localised
contracts localised
global standards
unnecessary
extends
edge-pricing
[Shenker96]
13
context
pricing
pricing
clearing
split edge pricing
scenarios
-
summary
per class
 each class
of service
has a
market
 class
mismatch
effects
localised
19 Nov 1999
Direction of Value
14
context
pricing
pricing
clearing
scenarios
split-edge pricing
Nd
Qv
Ne
Nd
Qu
Nb
V Qt
19 Nov 1999
Wbar
Wabs
Qu
per leg
 multicast &
heterogeneous
Qw
QoS
Qw see sister
paper for
mathematical
Qu model
 unicast,
multicast and
Nc
concast
Wbas
Wabr
Na
summary
Nc
Direction of Value
15
context
pricing
pricing
clearing
scenarios
summary
cost approximation per domain
 cf. weighted
star net
 cf. half circuit
charging
 covers costs on
average
 separate rcv &
send prices fit
asymmetric
costs
19 Nov 1999
Direction of Value
16
context
pricing
pricing
clearing
scenarios
service summary
data
money
interconnect example
-10
-10 10
-10
D
10 10 37
-1•1
11-8•4
10-10 10
-14
3
B
10 -10
E
17 10
-1•1
10 -10
-8•2 11
3
-14
+4-1•3
14
-11
14
-11
9-2•4 15
-12
A +11
-1•1
-8•1
12
-15
+25
-1•1
-7•3
-9
14
-14
19 Nov 1999
3
14
14
-14
-14 0
14
-9
12
-10
12
-10
10
-12
9 9
C
9
Surplus:
A: 2= 11 - 9
B: 1= 9 - 8
C: 1= 4 - 3
D: 4= 37 -33
E: 0= 17 -17
F: 3= 25 -22
G: 3= 18 -15
Tot:14=121-107
+18
-1•1
-7•2
10
2 2 -12
9
-9
F G
10
10
-10
10
-10
Costs:
Direction ofhalf
Value interconnect B=2, others=1
17
access F,G=7; A,D,E=8
context
pricing
pricing
clearing
scenarios
summary
default value apportionment
 choose default to minimise clearing
'only senders' or 'only receivers' charged
halve transaction volume?
requires industry-wide co-operation
unstable - mavericks gain and increase costs of all
multicast requires trusted receiver count [Holbrook99 etc]
 'both sender and receiver' charged
majority of comms between consenting parties
value, charge and cost congruent
stable against maverick policies
doesn't match blame for unicast transmission...
19 Nov 1999
Direction of Value
18
context
pricing
clearing
scenarios
summary
blame, liability & control: multicast
 joint blame for multicast is nice
join
join
sender
router 1
router 2
host 3
host 1
join
host 4
join
host 2
receivers
 but once receiver has joined…
– e.g. surprise high QoS multicast traffic
• should receivers pay for this?
 …problem same as for unicast…
19 Nov 1999
Direction of Value
19
context
pricing
clearing
scenarios
summary
blame, liability & control: unicast
 but…
...sender always to blame for a unicast or
for 'multicast surprises'
 other confusions clarified:
•
•
•
•
19 Nov 1999
request-reply is irrelevant to network layer
value of moving info transitory
value of moving GBs is often for a few bits
cost of xmt << cost of targeting what to xmt
Direction of Value
20
context
pricing
clearing
scenarios
summary
dilemma & solution
 'sender and receiver both charged'
vs. 'sender to blame'
 receiver customarily pays
...but...
sender liable
 cf. postage stamp
• sender customarily pays, but receiver liable
19 Nov 1999
Direction of Value
21
context
pricing
clearing
scenarios
summary
clearing across edges: third party
service
12
money
3
2
46
51
5
45
50
5
5
44
49
clearing
150
19 Nov 1999
Direction of Value
22
context
pricing
clearing
scenarios
summary
clearing across edges: iterative
service
7
money
3
44
46
100
19 Nov 1999
54
44
51
Direction of Value
24
context
pricing
clearing
scenarios
summary
end-to-end model: justification
avoids Int'l Accounting Rate System flaws:
muddling wholesale with clearing
per-flow metering in backbone
trusted end counting for multicast
apportionment needs authentication
therefore unaggregateable
O(acd+1) prices in system  6e26
d5
a = apportionment
schemes per ISP
 10
19 Nov 1999
Direction of Value
c  20,000 25
context
pricing
clearing
scenarios
summary
end-to-end model: justification II
 localised contract
• jurisdiction clear
• what to meter decided locally
• only need bulk backbone metering
 selective openness of clearing process
O(ac2) prices problem still exists  4e9
• but at least feasible
19 Nov 1999
Direction of Value
26
context
pricing
clearing
clearing
scenarios
summary
exception peering
 exceptional failures
– loss of a packet subject to reservation
– excess delay of a low latency packet
 no customer revenue
 no recriminations between ISPs
– no “who lost packet?”
– no “who delayed packet most?”
 assume peers sufficiently dimensioned
19 Nov 1999
Direction of Value
28
context
pricing
clearing
scenarios
scenarios
summary
scenarios
 see paper for...
 finding an end-to-end price
 sender liable but local payment customary
 inter-domain multicast with heterogeneous
quality of service (QoS)
 phone to Internet gateway (PIG)
 per session accounting  bill=anachronism
19 Nov 1999
Direction of Value
29
context
pricing
clearing
scenarios
scenarios
summary
PIG - phone to Internet gateways
PSTN
service
Internet
money
iterative
end-end
3
1
7
3
2
36
41
100
19 Nov 1999
5
21
5
clearing 59
29
25
24
21
PIG
Direction of Value
30
context
pricing
clearing
scenarios
summary
summary
connectionless value flow
 split-edge pricing scales infinitely
• price effects localised
• no need for global pricing, accounting, metering
standards
• no per flow metering away from edge
• no need for global contractual environment
• no such thing as trusted multicast receiver count?
 'sender and receiver both charged' stable
• customary for unicast - but only sender liable
• telco model must not leak across PIG
19 Nov 1999
Direction of Value
31
context
pricing
clearing
scenarios
summary
summary
alternative now seems obvious...
 value flows outwards
• if unequal value, clear end-to-end
• implications for design of multicast diffserv
 openness to commercial innovation
 end-to-end intermediary could take 0-50%
of retail ISP and telco markets
• ...and their customer relationships
19 Nov 1999
Direction of Value
32
context
pricing
clearing
scenarios
more
summary
info
further information
 Differential charging (diffchar) project
http://www.labs.bt.com/projects/mware/
 Bob Briscoe
http://www.labs.bt.com/people/briscorj/
19 Nov 1999
Direction of Value
33