Transcript MIND97 - Columbia University

Modest networking
Joint work with
Maria Papadopouli
Hannes Tschofenig
Xiaoming Fu
Jochen Eisl
Robert Hancock*
Henning Schulzrinne
Columbia University
MIND Workshop – London, Oct. 7, 2002
Keynote Address
(Opinions are the speaker’s and may not be shared by the co-authors…)
Overview




New Realism in networking
Multimodal networking
Some thoughts on QoS
CASP – another attempt at QoS
signaling
New realism

Old notion: build it and they will pay



Convergence  all bits are similar


“Don’t want to be just dumb bit carriers’’
Value add  charge exorbitant fees for
shipping special bits
Cheapest bits win (unless monopolies or
regulation interfere)
Almost all of networking is/will be a
commodity – we should be proud of it!
Cost of networking
Modality
mode speed
OC-3
P
155 Mb/s
$0.0013
Australian DSL
P
512/128
kb/s
$0.018
GSM voice
C
8 kb/s
$0.66-$1.70
HSCSD
C
20 kb/s
$2.06
GPRS
P
25 kb/s
$4-$10
Iridium
C
10 kb/s
$20
SMS
P
?
$62.50
P
8 kb/s
$133
videoconferencing or 1/3 MP3)
(512/128 kb/s)
(160 chars/message)
Motient
(BlackBerry)
$/MB (= 1 minute of 64 kb/s
Spectrum cost for 3G
Location
what
cost
UK
3G
$590/person
Germany
3G
$558/person
Italy
3G
$200/person
New York
Verizon
(20MHz)
$220/customer
Generally, license limited to 10-15 years
Multimodal networking




= use multiple types of networks, with
transparent movement of information
technical integration (IP)  access/business
integration (roaming)
variables: ubiquity, access speed, cost/bit, …
2G/3G: rely on value of ubiquity immediacy


but: demise of Iridium and other satellite efforts
similar to early wired Internet or some
international locations

e.g., Australia
Multimodal networking


expand reach by leveraging mobility
locality of data references


mobile Internet not for general research
Zipf distribution for multimedia content



short movies, MP3s, news, …
newspapers
local information (maps, schedules, traffic
radio, weather, tourist information)
Multimedia data access
modalities
bandwidth
(peak)
delay
high
low
high
7DS
802.11
hotspots
low
satellite
SMS?
voice (2G,
2.5G)
A family of access points
2G/3G
WLAN
hotspot + cache
7DS
Infostation
access sharing
7DS options


Many degrees of cooperation
server to client




only server shares data
no cooperation among clients
fixed and mobile information servers
peer-to-peer

data sharing and query forwarding among
peers
7DS options
Query Forwarding
FW query
query
Host A
Host B
Host C
time
Querying
active
(periodic)
passive
Power conservation
communication enabled
on
off
time
Dataholders (%) after 25 min
high transmission power
Dataholders (%)
100
P2P
90
80
P2P data sharing
(power cons.)
Mobile Info Server
70
P2P data sharing
60
50
P2P data sharing & FW
(power cons.)
Fixed Info Server
40
Fixed Info Server
30
20
10
Mobile Info Server
0
0
5
10
15
20
25
2
Density of hosts (#hosts/km )
Message relaying with 7DS
WLAN
Message
relaying
Host A
Host B
WAN
messages
WLAN
Host A
Gateway
Quality of Service
Why QoS hasn’t happened





need to admit failure – “bandwidth too cheap to meter”
undemocratic: some traffic is more equal than other
dishonest: we only talk about the beneficiaries
reminds you of your mom: no, you can’t have that 10 Mb/s now
socialist: administer scarcity - we like SUVs (or to drive 100
mph)!

“risky scheme”: security exposure – reserve your whole network
niche only: displacement applications (such as telephony) need

touchy-feely: requires cooperation  edge-ISP, transit ISPs, end

snake oil: add QoS, lose half your router interface bandwidth

QoS
systems
What makes QoS hard


No, it’s not RSVP scaling
network has become harder to evolve:






network address translation
firewalls
high packetization overhead (VPNs, IPv6)
nobody can be trusted
to be useful, has to be nearly universally supported
(“no, you can’t make calls to AS 123”)
network QoS vs. business class model: “coach is
empty, please refund fare”


almost all the time, reserved traffic gets same delay as best
effort
applications will switch QoS classes
What makes QoS hard

currently, the ISP interface is IP and BGP –
adding a third one is a big deal



new Internet service model: TCP client
(inside) – server (outside)


trust model  ISP or cash model
payment model completely unclear for peering
exception: peer-to-peer on college campuses
network to host: you first, no, you first
What is QoS, really?

Network transparency





no loss (< 1%)
very few delay spikes (< 1%)
close to propagation delay
anything else is too hard to explain to users!
QoS is just a facet of network reliability



consistent 5% packet loss is much better than 5%
probability that network is unavailable for seconds
users are willing to pay for availability
traditional QoS may help availability during outage periods



e.g., MCI/UUnet breakdown 10/3/02
DOS attacks
failure of load distribution links
CASP = Cross-Application
Signaling Protocol


RSVP is being used for lots of things beyond flow
setup: RSVP TE, midcom, …
Complex and monolithic

multicast support 






fairly closely ties QoS to RSVP


multiple reservation styles
killer reservations and error handling
receiver-orientation only
non-RSVP region handling
interface complexity (LIH)
hard to extract generic signaling protocols
 CASP as modular signaling protocol

currently a proposal for IETF NSIS group
What is CASP?

Generic signaling service


establishes state along path of data
one sender, typically one receiver




can be multiple receivers  multicast
can be used for QoS per-flow or per-class reservation
also: firewall setup, TE, programmable networks,
configuration, topology exploration, …
avoid restricting users of protocol (and religious
arguments):


sender vs. receiver orientation
more or less closely tied to data path


router-by-router
network (AS)
CASP network model – onpath
selective
CASP chain
QoS
QoS
QoS
midcom
omnivorous


CASP nodes form CASP chain
not every node processes all client protocols:



non-CASP node: regular router
omnivorous: processes all CASP messages
selective: bypassed by CASP messages with unknown client
protocols
CASP network model – out-ofpath
Bandwidth broker
NAC
CASP
AS15465
AS 1249
AS17
data
Also route network-by-network
 can combine router-by-router with outof-path messaging

CASP protocol structure
client layer
(C)

messaging layer
messaging layer
(M)
(M)
transport layer
UDP
(T)
IP router alert
client layer does the real
work:




scout protocol
reserve resources
open firewall ports
…
messaging layer:


establishes and tears down state
negotiates features and capabilities

CASP
transport layer:

reliable transport
Next-hop discovery

Next-in-path service







enhanced routing protocols  distribute information about
node capabilities in OSPF
routing protocol with probing
service discovery, e.g., SLP
first hop, e.g., router advertisements
DHCP
scout protocol
Next AS service



touch down once per autonomous system (AS)
new DNS name space: ASN.as.arpa, e.g., 17.as.arpa
use new DNS NAPTR and SRV for lookup

similar to SIP approach
Mobility and route changes
DEL (B=2)
discovers new route
B=1
on refresh
ADD
B=2



avoids session identification by end point addresses
avoid use of traffic selector as session identifier
remove dead branch
Conclusion



Until wireless bits are too cheap to
meter, try hiding lack of universal high
bit density from user
QoS is a reliability mechanism
CASP as a new signaling platform