Transcript ATM!!

Networking For the Future
By
Dr. Junaid Ahmed Zubairi
Sigma Xi Brown Bag Seminar
Oct 3rd, 2003 at 12 Noon
Overview of Presentation










The Evolution of Internet
The Types of Traffic on the Internet
Performance Issues in Packet Switching
The World Wide Web
Changing Traffic on the Internet
The Birth of ATM!!
ATM’s Service Classes and Layers
Intserv, Diffserv, MPLS, TE
GMPLS
Future Network
Fig 1: The Evolution of Internet: Past and PresFu
Seminar References
 Computer Networking: A Top Down Approach
Featuring the Internet by Kurose and Ross,
Addison Wesley 2001
 ATM With X-Cell, XYLAN Course 701, XYLAN
Inc.
 Computer Networks: A Systems Approach
Peterson and Davie, Morgan Kaufmann 2000
 Computer Networks Andrew Tanenbaum Prentice
Hall 1996
The Evolution of Internet
 DARPA (Defense Advanced Projects Research
Agency) funded the development of the
Internet.
 The first working network was ARPAnet that
was started in 1969 between four nodes
 The emphasis was on developing a robust
network that would continue to function even
if some of its parts were bombed out
The Evolution of Internet
 email,
 usenet,
 file transfer and
 remote login were the main Internet
applications
Types of Traffic on the Internet
 These applications generated almost
identical traffic stream on the Internet
 This traffic required “reliability”.
 The protocols were expected to deliver
all the data no matter how long it took
Performance Issues in Packet
Switching
 The TCP/IP suite of protocols breaks the
data into datagrams or packets and routes
each packet through an independently
selected path
 Packets may arrive at the destination out of
sequence but due to buffering and reordering, the actual data can be recovered
easily
Message Text
 Dear John
 I agree with your suggestions. I think we
may want to include Mike in the discussion.
I am forwarding this message to him as
well. He is an expert in this area and we
should wait for his comments before
proceeding forward on this project.
 Junaid
Message Broken Down
Dear John
I agree with your suggestions. I think we may want to
include Mike in the discussion. I am forwarding this
message to him as
Packet # 1
well. He is an expert in this area and we should wait for
his comments before proceeding forward on this
project.
Junaid
Packet # 2
Pkt#2
Pkt#1
Fig 2: Packets may follow longer paths and arrive
Pkt#2
Pkt#1
Fig 2: Packets may follow longer paths and arrive
Pkt#2
Pkt#1
Fig 2: Packets may follow longer paths and arrive
Performance Issues in Packet
Switching
 Selecting a path is called
routing and the
intermediate nodes from
source to destination are
called routers
 Each router builds up a
routing table to keep track
of reachable destinations
 If more than one path is
open to destination, the
router may select the
“best” path
Performance Issues in Packet
Switching
 Path selection criterion is usually shortest
path first
 If the shortest path is congested or
unreliable, the router can choose another
path
 The traffic is bursty and it can increase or
decrease abruptly based on the way the
Internet is used
Performance Issues in Packet
Switching
 Given this scenario, a router may find itself
overwhelmed with a lot more packets than it
can handle
 Usually routers would use simple FIFO
scheme to select the next packet to be
transmitted from a queue of packets
Router Exposed
Performance Issues in Packet
Switching
 If the queue is full, the newly arrived packets must
be dropped (or discarded)
 Thus increase in traffic may increase time-outs,
retransmissions and decrease in efficiency
 Usually, congestion in the network results in
delay and loss penalties
 Congestion builds up due to bursty users, no
active resource allocation and selfish users trying
to monopolize the bandwidth
Performance Issues in Packet
Switching
 Traditional TCP/IP based Internet can be
described as
–
–
–
–
“Best Effort”
“One Size Fits All”
“Hardly Any Service”
“World Wide Wait”
The World Wide Web
 Web deployment is flexible and easy
 Due to the web technologies, the Internet
has been put to use in almost all areas of
human knowledge
 For example, water distribution monitoring,
real-time traffic maps of big cities, free long
distance calling, distance learning with
lecture videos, buying and selling shares,
online shopping etc., the list appears endless
The Changing Traffic on the
Internet
 Due to the web enabled applications on the
Internet, there has been a tremendous
change in the types of traffic
 Now we have to deal with a significant
amount of traffic that is time-sensitive
 For example, consider the case of an audio
based application that needs to transmit the
data across the Internet
The Changing Traffic on the
Internet
The Changing Traffic on the
Internet
The Birth of ATM!!
 As the users started to use the web for time-
sensitive applications, they did not get what
they wanted--- a consistent acceptable
performance
 On some occasions, the network would give
the best performance and on other
occasions, it would be horrible in terms of
delays and lost packets
The Birth of ATM!!
 In this scenario, ATM offered a great
promise to the users
 ATM standards started taking shape in mid1980’s as telcos pushed for integrating
voice, video and data networks
 ATM was developed with the right targets
and in mid-1990’s, it offered the much
awaited performance assurance
What is ATM?
ATM:
 Is a cell-switching and multiplexing
technology that combines the benefits of
Circuit Switching (consistent transmission
delay and guaranteed bandwidth) with those
of Packet Switching (flexibility and efficiency
for intermittent traffic).
Why is ATM needed?
Switch
 Need to mix data, voice, and video traffic.
 We cannot just throw more bandwidth at the
problem
Why is ATM needed?
“A”
Switch
“C”
“B”
 All data packets are fragmented into fixed size cells
 Segmentation & re-assembly only occurs at end stations
 Time critical traffic on segment “A” only has to wait for the
current cell of “B”s data packet to be sent before it can get the
wire and be transmitted
 The ability to interleave cells from different messages is
instrumental to the operation of ATMs QoS.
ATM Cell
Header
Payload
5 bytes
48 bytes
 Small Cells - 53 bytes long
– 5 byte header
– 48 byte payload
 Fixed Length = Fast Switching
 Fixed Length = Contracts can be established and QoS
maintained
AAL Types
User Traffic: Voice, video and data
ATM Adaptation layer
ATM layer
Physical layer
AAL1 is for circuit emulation
– Class A - constant bit rate and time sensitive traffic
AAL5 is for compressed video and data (used in IP over ATM)
– Class B - variable bit rate and time sensitive traffic
– Class C - variable bit rate (e.g., Frame Relay)
–
Class D - variable bit rate, connectionless
Service Categories
 Categories are based on type of traffic and type of
service
–
–
–
–
–
CBR
rt-VBR
nrt-VBR
ABR
UBR
Constant Bit Rate
Real-Time Variable Bit Rate
Non-Real Time Variable Bit Rate
Available Bit Rate
Unspecified Bit Rate
-Voice
-Video
-Frame Relay
-Data
-Data
Traffic Management
Traffic Management
 Two opposing views
– Enforce several rules in order to regulate the
traffic and adapt to the available bandwidth
– Add more bandwidth
 For example, highways enforce HOV rule,
speed limits, traffic light controlled ramps
to enter the highway etc. OR autobahns
with no such rules
CAC
I want to send
traffic of this
type, and want
this QoS
CAC
Can I support this
reliably without
jeopardizing other
contracts
Guaranteed QoS request
No or Yes,
Agree to a
Traffic Contract
ATM Network
• If CAC passes, network and
user agree on a traffic contract
VPIs and VCIs
 Virtual Path Identifier (VPI)
 Virtual Circuit Identifier (VCI)
OC-3 155 Mbps
VPI-10, 50 Mbps
VCI-100
VCI-101
VCI-100
VCI-101
VPI-20, 60 Mbps
Remaining 45 Mbps is in VPI-0
Traffic Shaping and Policing
 In order to meet the QoS contract
obligations, ATM network enforces traffic
shaping and policing
 Shaping involves techniques such as “Leaky
Bucket Algorithm” to regulate bursty traffic
 Policing means marking CLP (Cell Loss
Priority) on the offending cells that violate
the maximum rates agreed
Leaky Bucket Algorithm
Why ATM Failed!!
 ATM failed because of several factors
– ATM is too complex (From packets to cells to SONET
frames, using AAL’s, emulating LAN)
– ATM is expensive
– Ethernet has evolved into much faster 100Mbps and
1000Mbps services
– All popular and established network applications are
packet based
 ATM is down but not out. It is used in the telco
cores
Life after ATM
 IETF (Internet Engineering Task Force) is a
very large organization with thousands of
members
 IETF identified the problems with the
traditional Internet and engaged in a long
and continuing effort to improve the
services and management
Quality of Service
 New applications need performance and
resource assurance
 Service differentiation is also needed so that
the traffic from different applications is
treated in service-appropriate way
 Resource assurance and service
differentiation means QoS (Quality of
Service)
IETF’s Models
 Targets:
– Internet should be run in a way that there is no
congestion
– Applications should be able to reserve or obtain
network resources at a given QoS
 IETF has been working on developing new
models and protocols for the Internet
 During the last decade, Intserv and Diffserv
models have been developed
Integrated Services
 Intserv stands for “Integrated Services” and
requires reservations before transmission
 To receive resource reservation, an
application describes its requirements
 The network determines a path based on the
request
Reservations, Reservations,
Reservations
Intserv
 A reservation protocol is used to install the
reservation state along the selected path
 The reservation setup protocol in the Intserv
model is the RSVP (Resource ReSerVation
Protocol)
RSVP’s Services
 RSVP offers two types of services
 CONTROLLED LOAD service means that
the service offered to a flow in an
overloaded network is the same as it would
get in a lightly loaded network
 GUARANTEED SERVICE is when a flow
gets hard guarantees on the delay it will
suffer
RSVP Problems
 RSVP relies on extensive signaling for obtaining
flow reservations along a path.It also entails soft
state overhead and therefore does not scale well to
the Internet
 Most of the Internet traffic consists of short-lived
web transactions. It will be unwise to go through
reservations for such traffic
 RSVP may be successfully deployed in a campus
network but not on the global network
IETF’s DiffServ Model
 Intserv’s problems prevented its deployment
 IETF started developing a new model in 1997 to
provide differing levels of service to different
applications without the overhead of signaling and
state maintenance
 The DiffServ model uses the TOS field in IPv4
header to affix labels on packets belonging to
different service levels
 DiffServ has the potential to offer QoS on the
Internet, at last!!
IETF’s DiffServ Model
 Consider a gas station, you can buy regular,
super or premium gasoline from the same
pump
 DiffServ offers various service levels to the
customer from the same network with SLA
 DiffServ adopts techniques used in ATM for
traffic management, in a simplified way
No Reservations Ever!!
VIP
Diffserv Outline
 Diffserv works on the basis of dividing the
traffic into a small number of forwarding
classes
 For each FEC, the amount of traffic entering
the network is controlled at the edge of the
Diffserv network
 FEC’s are prioritized, with each one coded
into the IP header’s TOS byte. Core routers
offer priority treatment based on the coding
Diffserv Edge Router Functions
Per-Hop Behaviors
 IETF has defined two DS services that are
visible as PHB (per-hop-behavior) of an
intermediate router for the marked packet
 EF (Expedited Forwarding)
– EF is the premium service offered. It can
appear as a virtual leased line for the customer.
It offers low loss/latency and assured
bandwidth
 http://www.ietf.org/rfc/rfc2598.txt
Per-Hop Behaviors
 AF (Assured Forwarding)
– The AF PHB group provides delivery of IP
packets in four independently forwarded AF
classes. Within each AF class, an IP packet can
be assigned one of three different levels of drop
precedence. A DS node does not reorder IP
packets of the same microflow if they belong to
the same AF class.
 http://www.ietf.org/rfc/rfc2597.txt
Hybrid Approach
MPLS
 MPLS was developed to map IP over ATM
because the core routers use ATM. MPLS
has additional features that are too exciting
to ignore
 In MPLS, a short fixed length label is
encoded into the packet
 The intermediate LSR (Label Switched
Router) finds the next hop from a table,
using the label as an index
MPLS
 If the LSR is an ATM switch, label is just
the VPI/VCI identifier
 If the LSR is an IP router, the label
eliminates the destination based routing and
reduces the router to a label switch
 A label switched path (LSP) must be set up
prior to the start of transmission
LSP Hierarchy
LSP’s in an MPLS Network
MPLS & TE
 Instead of routing, now the routers do label
switching, a much faster job
 Network manager can decide LSP’s (label
switched paths) based on load distribution
and other administrative goals
 Directing traffic on paths not determined by
traditional IGP’s provides flexibility and
load balancing. It is known as TE (Traffic
Engineering)
MPLS AND TE
 MPLS runs constrained routing to
determine an LSP within an MPLS domain.
 LSP may have some QoS features, based on
the algorithm used
 The path could be strictly specified or
loosely outlined and backup paths may be
specified for handling link failures
Automated Provisioning
 The networks are growing bigger!!
 The protocols are becoming more complex
 With Diffserv, MPLS, RSVP-TE, CR-LDP,
COPS and associated protocols, it is
impossible to allow manual provisioning
 Therefore, there is a need for automated TEbased path selection algorithms
QoS Traffic Considerations
 If only the available bandwidth is
considered, the class of service may not be
taken into consideration
 Thus, the best effort traffic may intersect the
QoS traffic at several points within the
domain
 In Diffserv, this may be a recipe for
disaster!!
TELIC
 An efficient dynamic traffic engineering
algorithm is developed for selecting paths
across an MPLS-Diffserv domain
 TELIC (Traffic Engineering with Link
Coloring) works with a set of traffic
requests present at an ingress router of such
a domain
 It allocates paths to an egress node using
Dijkstra’s shortest path algorithm
TELIC
 Each request specifies the amount of
bandwidth requested followed by the
Diffserv class of service (EF,AF,DF)
 While processing a request, TELIC
partitions the network into several
monochromatic subgraphs and makes an
effort to match the request with an
appropriate subgraph
TELIC
 In case a subgraph has no path to the egress
node, TELIC merges it with another
subgraph as per rules carefully built-in and
starts the search all over again
 In case a search is exhausted, rules are
available to deallocate a best effort class
LSP and start the search again
 TELIC is written as a flexible tool in C++
5=
BL
20
1=
L
B
1
BL
BL2=6
5
White
65
0=
Router 4
BL
15
=8
0
BL16=20
BL11=100
Router 8
Router 5
0
BL13=60
Router 6
Silver
BL
Router 10
Router 9
4=
40
40
7=
L
B
Green
Yellow
Router 7
65
9=
BL
0
=4
12
BL
BL
3=
6
BL6
=80
5
Router 0
BL
8=
40
BL
17
=6
0
Router 1
Router 3
40
6
4=
L1
B
All possible paths
P0: 0 1 3 10
P1: 0 1 4 7 10
P2: 0 1 4 8 10
P3: 0 2 9 8 10
P4: 0 2 9 10
P5: 0 5 4 7 10
P6: 0 5 4 8 10
P7: 0 5 8 10
P8: 0 6 9 8 10
P9: 0 6 9 10
Router 2
Red
Figure 6: An MPLS domain
GMPLS
 The Internet backbone must use optical switching
instead of electronic switching to handle the
projected huge bandwidth
 MPLS cannot handle non-packet switching
 Recently the industry has gravitated towards
GMPLS (Generalized MPLS) as the control plane
solution for automatic lightpath setup and
teardown in optical networks
 GMPLS is an extension of MPLS
 GMPLS allows control and provisioning of nonpacket devices
Why Optical Networking?
(Courtesy Prof. Raj Jain Ohio State University)
GMPLS Layers (Courtesy Prof. Raj Jain Ohio
State University)
GMPLS
 Using GMPLS, it is possible to perform
switching based on:
–
–
–
–
–
Wavelengths
Wavebands
Timeslots
Ports
And Labels
GMPLS
 For example, in an all-optical switch, there
may be thousands of tiny mirrors that can
be moved by miniature motors
 Switching can be done by adjusting a mirror
so that light entering from one fiber can be
reflected (switched) to the desired path
forward
LMP
 A link management protocol has been developed
for GMPLS. It provides link provisioning, fault
isolation and link aggregation
 Selection of label in MPLS  Selection of
wavelength and OXC port in GMPLS
 MPLS LSP  GMPLS lightpath
 Before GMPLS, control and provisioning of
optical network could take weeks!!
 Vendors were also reluctant to de-provision due to
any changes
End to End Provisioning
Ubiquitous Networking
 The future of networking is being defined
today. It is planned as a global network
with no breaks or bumps
 Users may roam around with notebooks and
remain connected wherever they go!!
 The realization of this goal calls for a global
wireless network, global wired network and
an interface between the wired and wireless
networks
Mobile Networking
 Mobile networking is developed rapidly
with IEEE 802.11and Bluetooth standards
 Let us take a look at the various
configurations possible with IEEE 802.11
Wireless LAN with access
point under IEEE 802.11
Ad-hoc network
Issues
 Several issues are being investigated for
improving Wireless LAN functionality
 For example, how to perform transparent
handoff ?
 How to perform routing in an ad-hoc
network?
 How to shape and mark the traffic to esnure
good QoS?
Bluetooth
 Bluetooth is more focused on connecting
electronic gadgets like digital cameras,
mobile phones, printers, mouse etc. with
each other and with the computer
 It replaces infrared line-of-sight type of
connection
 It faces tough competition from UWB (ultra
wideband), a similar technology but much
faster than Bluetooth
Future Network
 We envision a global ubiquitous network with
instant access to email and the web from anywhere
in the world
 With tremendous capacity, the network would
offer the desired quality of service to our
multimedia applications
 Traditional phone network will become a tiny
section of the overall Internet
 Videophone over the Internet is expected to be the
next killer app
 Privacy?? Huh