Packet Data Evolution S. Wood Nov. 2006 Copyright 2006 Modern Systems Research

Packet Data Evolution

S. Wood Nov. 2006

Copyright 2006 Modern Systems Research

Networking Local Area Networks

• Became popular with LAN’s (Ethernet @ 10Mbit)• How do you connect 2 LAN’s together?• Many LAN’s??• High Speed?? (100Mbit)• Internet??• Networking Software

DEC, 3COM, Banyon, Novel, Microsoft• Bridges, Switches, Routers


Internet

• Uses IP routing

• No support for QOS– QOS assigns priority to certain payloads

• Packet traffic “Bursty” by nature

• Congestion often occurs

• Congestion causes packet loss / delays

• Higher level protocols provide restoration of lost & damaged packets


Inside the Internet

• User pays service provider for service

• Service provider pays higher level service provider

• Peering arrangements:– service providers agree to “share” access to users

• Internet Routing


IP Routing

• IP (v.4) Address = 32 bits, 4 bytes• Static Hierarchal Routing Class A (256 sub-address Class Bs) Class B (256 sub-address Class Cs) Class C (256 Users)

Users (256)Addresses assigned based on agreements between service providers• Subnet mask on router tells it what block or part of block to look at• Dynamic Routing: users may move around requiring the network to

“look” for a user• Routing Protocols allows routers to communicate to find the best path

to forward packets


Routing Protocols

• Allows routers to work together to find the best path to a user or a group of users

• Protocol examples:– OSPF, RIP, BGP ,IGP, RSVP

• Common attributes:– Use Static parameters to calculate route example: latency based on distance– Some protocols are “link state” Flood network with Link State Advertisements (LSA)

• Routing protocols do NOT take into account dynamic parameters


Dynamic Parameters

• Congestion

• Intermittent Nodes & Links

• Packet flows

• QoS

• Policies

• Packet loading


Congestion control

• Most internet problems relate to congestionSome Solutions:• Network Traffic Engineering:

– identify the source of the congestion – provide additional bandwidth to fix specific problem.

• Identify heavy users (e.g. Gamers) and:– get them to pay for more bandwidth– apply flow constrictors

• Over provision bandwidth (again and again)• QOS based routing• Apply admission controls

– “Block new traffic until congestion subsides”


Packet Quality of Service

• QoS – Quality of Service– QoS based routing has been with us for several years– diff-serv (Differenciated Services) is a QoS Architecture

• Some QoS routers use:– “Weighted Fair Queuing (WFQ)”

• Congest means there are too many packets to be transmitted over a given path.

• WFQ provides an orderly means for discarding packets to bring the bandwidth down to what can be transmitted. All sources lose some packets with WFQ. Some sources lose more packets than others


Packet QoS (cont)

• Hard QoS:– Packets are organized into flows– Each flow is assigned a class of service– When congestion occurs, bottom classes are discarded

first. Higher classes are unaffected• Currently 4 classes have been identified

– Voice, Video, Priority data, Best effort

• Today's routers assign QoS based on the port used. All packets arriving at that port share the same class of service.

• Next generation routers assign QoS based on packet flowsCopyright 2006 Modern Systems Research

IP Routing – Open Issues

• Currently includes no QOS guarantees• IP routing software can crash under heavy

congestion• Latency not deterministic• No admission control• No ability to perform load balancing• Internet routing protocols do not take into account

dynamic parameters such as congestion• Network slow to recognize link or node failures


Internet Today

• No QoS• Congestion takes down network• Variable Latency• No TDM Voice• Network Traffic Engineering needed• Not Secure

Standard Router

Packet Flows

PacketData

Network

TCPUDP

IP

L4 L3


SecurityInternet Security Threats

• Denial of service attacks• Phishing• Spam• Mal ware• Viruses and wormsIP networks use in band signaling• Internet vulnerable to new generation of “Blue Boxes”• Users can modify internet packets to:

– Upgrade QOS– Spoof source / destination– Modify address at will

Next Generation Routing• New routers are being developed to correct the issues in current

routing protocols• Flow-based routing:

– Packet flows are groups of similar packets traveling together sequentially

• Examples:– Voice, Streaming Video, File transport

• Flow based routers must:– Sort packets into flows– Buffer each flow– Implement policy/QoS rules to each flow– Forward packets

• Examples of companies building next generation routers:– Caspian Networks, Anagran


Next Generation Routing:The Hard Part

• When IP addressed packets arrive at the router, they all look the same. The router must move up the protocol stack to layer 4 sort the packets into flows (Deep Packet Analysis).

• Once sorted, The flows must be ordered and assigned QoS and policy constraints.

• When done properly, the network can give priority to voice or video when congestion occurs. Some technologies such as MPLS already support QoS


Internet With A Flow Based Router

• Expensive (Flow Based Routers needed throughout the network)

• Congestion can still take down Packet Switches (must be Traffic engineered)

• Does not support TDM voice• Not secure

Flow BasedRouter

Packet Flows

Flow BasedPacketData

Network

TCPUDP

IP

L4 L3


Label SwitchingTDM Voice

• Used in DACS (1980’s) – DACS is a digital access cross connect– Uses a time slot interchanger– Each DS0 (64kb data channel) can be considered as a

packet; it has both a payload (8 bits) and an address (from the framing)

– The old address is used to reference the new address in the cross connect.

– Path and setup software ran separately from the cross connect hardware


Label SwitchingFrame Relay (early 1980’s)

• WAN Data Network service offered by telecommunications carriers

• Desirable, since it was tarriffed for less money than voice services

• Typically used T1’s• Each pack was variable length and had a short

label• Used bits to encode priority on each packet

(congestion control)• Slower and less efficient than later IP routing


Label SwitchingATM (mid 1990’s)

• Stands for Asynchronous Transfer Mode• Considered next level for ISDN• Supported both Voice and several data formats

– AAL1 through AAL5• Uses 53 byte cells (5 byte address, 48 byte payload)• Intended to be used on SONET• Fundamental Switching very simple, Transport and setup

become complex• Supports hard QOS. Switching very reliable. Adopted by

most carriers• Went out of favor with the introduction of gigabit Ethernet


Label SwitchingMPLS

• Stands for Multi Protocol Label Switching• 4 byte label, variable length payload• Used with SONET up to 40 Gbit/S• Considered very reliable since switching is

done by hardware• Uses internet routing protocols• Considered next transport technology for

carriersCopyright 2006 Modern Systems Research

What is MPLS?• MPLS: Multi-Protocol Label Switching• A short 20 bit label is attached to the front of each packet.

A label is good for one hop only!• When a packet is forwarded by a Label Switch Router

(LSR), the next hop is assigned and the label is updated.• LSR’s are very fast, some even operate at 40Gbit/sec rate.• Labels can be stacked; A MPLS packet can have it’s label

stack “pushed” or “popped” instead of simply being translated at each LSR.

• The path taken by the labeled packets is called a “label Switched Path” or LSP

• The path is fixed and can traverse several nodes.


LSP Programming• Label Switched Paths are usually built from the

destination to the source.• A special protocol called LDP (Label Distribution

Protocol) exists for this purpose.• IP routing protocols are used to determine the best

path and build the LSP.• Building LSP’s this way can be problematical and

slow.• Many carriers choose to use network Traffic

Engineering to build and manage LSP’sCopyright 2006 Modern Systems Research

What is Network Traffic Engineering?

• Network Traffic Engineering is the computer simulation of a data network in order to ensure that congestion is avoided and the best links are chosen to carry the data flows between nodes.

• Part of Network Traffic Engineering is to set up redundant paths if a priority path was to fail


Why do Network Traffic Engineering?

• Large carriers have found that engineered networks are more efficient and robust

• Most carriers use MPLS on backbone systems.• MPLS works best when engineered.• The internet bubble generated a glut of bandwidth.

Consequently only a minor amount of engineering is needed.

• Automatic path allocation software supplied by vendors does not do an adequate job.

• Now that extra bandwidth has been used up and engineering must be done to avoid network crashes


What is Done?

• Network Traffic Engineering is often done by outside firms.

• Special network simulation software has been developed for this purpose.

• Network engineers collect data from carrier on nodes, links, policy preferences, existing LSP’s, estimated bandwidth needs for each LSP, latency between nodes, customer contract requirements, etc.

• Engineers enter collected data into simulation program.• Program produces a map with traffic loading and latency

for all nodes


What is Done? (cont)• A maximim traffic loading value is assigned to each node

(usually 70%)• Engineers use program to build new LSP’s or change

existing LSP’s to conform to latency and traffic requirements

• Stress tests are performed on simulated network to find failures and maximum loading

• Redundant LSP’s are added as a result of the previous tests.

• When finished, recommendations are made to carriers• Carriers provision new LSP’s


Internet Reliability• Internet reliability is linked to each router• Router Reliability (MTBF)

– Computer Logic 100,000 hrs– Power Supply 200,000 hrs– Routing Software 10,000 hrs– AC power 40,000 hrs

• Clearly, the routing software reliability is the weakest link• This is evidenced by the system “crashes” that occur when the

network is overloaded. The network relies on the routing protocols to bypass effected routers. Because of the delays involved, these crashes will cause dropped VOIP calls

• MPLS switches are more reliable because the actual switching is done in hardware.

• Adding QoS can worsen the reliability as it can slow the software making it easier to crash.


Internet Availability• An Internet outage is measured in time. This outage is often shown as

Mean Time To Repair or MTTR.• 1 – MTTR = Internet Availability generally expressed as a percentage.• The availability requirement for the PSTN is that it is available

99.999% of the time. It can only be out of service a total of 5.25 minutes a year!

• To achieve this Telcos had to do several improvements over traditional designs over time:– Redundant Power (dual 48V battery plants)– Transmission tests through switch matrix before call cut-through– Redundant processor and software throughout– Verification of connection paths before setup– Mechanized Loop Testing (MLT)– All links redundant with hot standby– Triple Redundant Switching Control Processor for SS7 network

• As can be seen, we have a long way to go before we abandon the PSTN in favor of whatever appears to be cheaper!


Toward Reliable VOIP

BatteryBackup

RedundantPower

DCpower

TraditionalAC Power

Heat

Power

Power

Switching

Switching

SoftwareSwitching

Switch usingHardware

RedundantProcessing

LabelSwitching(MPLS)(ATM)

Network

Network

FastRerout

PathVerification

ExcessBandwidth

RedundantLilnks

Congestion Linkfailures

Manyhops

Routingsoftware

Qualilty

Quality

HardQOS

FlowBased

Routing

AdmissionControl

NetworkTraffic

Engineering

SoftwareQOS

RouteFlapping

WeightedFair

Queuing

VOIP

PSTNQUALITY


Next Gen NetworkDesign Objectives

• Make network capable of carrying all forms of data– TDM Voice– High Speed Streaming Video– Internet traffic with QoS

• Make switching function hardware-based– Speed– Reliability– Deterministic throughtput

• Include path building and control in switching functions• Provide a simple request – grant (layer 4) user interface• Make a mesh network immune to node and link failures


Next Gen Network Design Objectives (cont)

• Network Plug and Play– Network can operate with no provisioning– Provisioning can be added to establish policies and QoS

• Self route each path based on :– Packet flow requirements: (latency, QoS, bandwidth)– Available nodes / links that meet the requirements– Service provider policies

• Support Enhanced Services– Multicasting (Branching & Merging)– Redundant paths

• OA&M Support (Operation, Administration and Maintenance)– Direct control of network setup by network engineers– Real-time network stress testing– Control alarms for node, link, path failures or congestion


Documents

Packet Data Evolution S. Wood Nov. 2006 Copyright 2006 Modern Systems Research