Fault Recovery Routing: High penalty cases and how to avoid them
Download
Report
Transcript Fault Recovery Routing: High penalty cases and how to avoid them
Internet Quality of Service (QoS)
issues in competitive commercial
network operation
Bjørn Jæger
Talk on given topic for the
Dr. Scient. dissertation
September 4, 2000
Department of Informatics
University of Bergen
Overview
Introduction
Internet
–
–
–
–
Quality of Service (QoS)
Overlay networks
Integrated Services
Differentiated Services
Multi Protocol Label Switching
Competitive
Summary
Jæger: given topic 2000
network operation
Internet development milestones
1876
Humans communicate by phone
1950’s
Computers communicate
First computer network: SAGE Semi-Automatic Ground Equipment (MIT, USA)
1969
ARPANET: Advanced Research Projects Agency NETwork
First network to use layered protocols, flow control, and fault-tolerance.
Introduced the term “Packet” for units of data sent: Packet Switched Network
1973
First ARPANET connection outside US
Direct Link: Virginia - Kjeller, Norway
1983
ARPANET starts using TCP/IP protocol connecting different networks
Became known as the Internet.
1986
IETF: Internet Engineering Task Force established
Open international community developing Internet standards
Now developing QoS related standards.
Jæger: given topic 2000
Internet development milestones
1990’s
Internet has become a world-wide data network
1991
1. Commercial Internet Operation
(NSF lifts restrictions on the commercial use)
2. World Wide Web (HTTP, HTML)
1992
First Internet audiocast (radio)
1993
First Internet video conference
1996
Internet phones catch the attention of US telecommunication
companies who ask the US Congress to ban the technology
(which has been around for years)
2000
High Demand for QoS sensitive services. Internet Service Providers
(ISPs) want to provide Voice, Video and Data over the Internet in
order to maximize revenue. Require new technology.
Jæger: given topic 2000
Traffic Handling in
the Present-Day Internet
Each
node broadcast Link State information
– Part of IGP: Interior Gateway Protocol
Administrative Domain
Each
node obtains a complete picture of all links
and routers (topology information)
Aggregation of states over Administrative Domains
Jæger: given topic 2000
Traffic Handling in
the Present-Day Internet
Each
router use the topology information to compute
the Shortest Path to every destination in the network
Molde-Ålesund:
Molde-Kristiansund:
Molde-Trondheim:
Molde-Bergen:
Molde-Oslo:
Molde-Ålesund
Molde-Kristiansund
Molde-Trondheim
Molde-Trondheim-Bergen
Molde-Trondheim-Oslo
T
K
M
Å
B
O
Based on this each router builds a Forwarding Table
associating an address prefix with the next hop link
Molde-Bergen Prefix:
…
Jæger: given topic 2000
Use link Molde-Trondheim
...
Traffic Handling in
the Present-Day Internet
When
a packet arrives at a router the Forwarding
Table is consulted, and packets are forwarded out on
the appropriate link based on the destination Internet
Protocol (IP) address
Each router makes an autonomous decision about how
to forward a packet
Forwarding proceeds in a connectionless way at every
hop
Jæger: given topic 2000
Problems with
the Present-Day Internet
Can
cause imbalance of network load since it use
the shortest path
T
Alternative paths:
K
congested
M
underutilized
Å
B
O
It is desirable to have routing which optimize the
traffic distribution for a given network topology
Jæger: given topic 2000
Problems with
the Present-Day Internet
No Traffic Management standardized: A router handles all
packets in the same way as fast as it can: Best Effort Service
Consequently, there is no way to predict a priori or guarantee
the Quality of Service (QoS) that a particular flow will receive
Voice, Video, and other real time services will not meet users
expectation of quality
Also, in case of link failures, link state is distributed
slowly: Slow convergence of routing to avoid failed device.
Need to be able to control the network resources in order to
provide a specified Quality of Service (QoS)
Jæger: given topic 2000
What is Quality of Service (QoS)?
Quality of Service can be characterized by:
–
–
–
–
Bandwidth (bits per second)
Delay: End-to-End
Packet Loss probability
Jitter (variation in delay among packets)
Human ear and eye are sensitive to delay and
delay variation (Voice & Video services)
Data applications are sensitive to data loss
Generally: Unequal allocation of resources among
connections are needed to provide QoS
Mechanisms: Routing, Classifying, Scheduling, Queuing,
Admission Control, Policing, Capacity Planning
Jæger: given topic 2000
Approaches to provide Internet QoS
Overlay
networks
Integrated
Services
Differentiated
MPLS
Jæger: given topic 2000
Services
The Overlay Solution
Run IP traffic over a circuit switched network, like e.g. ATM
Connect IP backbone by a complete mesh of permanent virtual circuits
ATM Core
ATM PVCs
Physical network topology
- IP over ATM
Jæger: given topic 2000
Logical network topology (mesh)
- PVCs serve as Point-to-Point IP links
The Overlay Solution
Advantages:
– Layer 2 (ATM) manage bandwidth
– Mesh of VCs: prevents hop-by-hop aggregation
– IP traffic can be individually routed through the layer 2 (ATM) topology.
– Can move traffic from overloaded links to underutilized links
Disadvantages:
– Need to build and manage two networks with different technologies
– Increased complexity of design and management
– Inefficient due to packing and encapsulation overhead associated with
Layer 2 (ATM)
Jæger: given topic 2000
The Overlay Solution
Disadvantages continued:
ATM PVCs
An edge router represents a User
Domain (many connections)
Problem: How to do the mapping between IP-connections and the PVC’s:
Is IP-connections aggregated by the user domain before transmission on a PVC?
Or
Is IP-connections routed over PVCs which are aggregated in the ATM-network?
Jæger: given topic 2000
Integrated Services - IntServ
Applications
set up paths and reserve resources
before any user data is sent
Four
–
–
–
–
components:
Signaling
Admission control
Classification
Scheduling
Jæger: given topic 2000
Integrated Services - IntServ
Signaling:
Resource ReSerVation Protocol - RSVP:
Carries resource reservation requests through the network with
characteristics of source traffic and QoS specifications required by
receiver (receiver oriented)
PATH w/Traffic Spec
Sender
PATH w/Traffic Spec
Receiver
Network
RESV w/QoS Spec
RESV w/QoS Spec
Available Resources
Admission
Control:
– Intermediate routers can reject or accept reservations
– Accept: Each router installs Flow State information
Jæger: given topic 2000
Integrated Services - IntServ
Classification:
– Upon receiving a packet each router perform a
classification based on Flow State, and places the
packet in a specific queue
Scheduling:
– The scheduler will schedule the packet for transmission
according to its QoS requirements
Jæger: given topic 2000
Problems with IntServ
Flow State information needed for each flow in
each router
Huge memory and processing capabilities in core
routers needed ==> does not scale
Requirement on routers are high, all need:
– RSVP, admission control, classification, scheduling
Jæger: given topic 2000
Differentiated Services - DiffServ
The IPv4 header:
Ver
4b
HdrLen Precedence
4b
3b
Type of Service
Unused
4b
1b
Total Length
16 b
Previously: Precedence bits used to indicate
– low-delay, high throughput, low loss service, but Not Standardized
IETF DiffServ working Group Charter:
– Define a set of classes by defining the layout of IPv4 DS-field
– Define packet forwarding rules: Per-Hop-Behavior, PHB
Jæger: given topic 2000
Differentiated Services - DiffServ
By marking DS-fields of packets differently and
handling packets based on their DS-field several
differentiated service classes can be created
Relative Priority Scheme
A customer get Differentiated Service from an Internet
Service Provider (ISP) by getting a Service Level
Agreement (SLA) with the provider
– Static SLA long term negotiation (months)
– Dynamic SLA needs signaling to request services on
demand (e.g. RSVP)
Jæger: given topic 2000
Differentiated Services - DiffServ
Customers mark DS-field of each packet
Classification, policing, shaping:
– At the ingress of an ISP domain (at edge) based on the SLA
– In core: behavioral aggregate
Service is allocated by granularity of a class
==> scaleable
Core
routers must be simple and fast, boundary
routers need not forward packets very quickly
since customer links are relatively slow
Jæger: given topic 2000
Problems with DiffServ
Per-Hop-Behavior (PHB) / Relative Priorities
Designing end-to-end services with weighted guarantees at
individual hops is difficult
Can not ensure resource availability inside the network
– based on assumption that arrival rate of Guaranteed Service is far
below the service rate - can not hold in general
– statistical guarantees only
Guarantees require stability of paths -- route pinning
needed
Jæger: given topic 2000
Multi Protocol Label Switching - MPLS
A label is inserted in IP packets at the ingress of an MPLScapable domain
– Mapping between IP-packets and labels is described by Forwarding
Equivalence Classes (FECs). Done only once, at the ingress.
IP-network
MPLS-domain
LSR
LSR
MPLS capable router examines only the label when
forwarding a packet: Label Switched Router (LSR)
Before packet leaves MPLS domain its label is removed
Jæger: given topic 2000
Multi Protocol Label Switching - MPLS
Labels are distributed by a signaling protocol
– Label Distribution Protocol (LDP) or RSVP
Explicit Route decided by source possible
– IETF work to extend IGP Link State information w/link capacity
– Facilitates Traffic Management
When a packet arrives at a router the label is used as
an index into the forwarding table which specifies the
QoS ==> FAST
The incoming label is swapped with the outgoing
label and the packet is switched to the next LSR.
Jæger: given topic 2000
Multi Protocol Label Switching - MPLS
Can
provide Guaranteed Quality of Service (QoS)
– Explicit Routes & Traffic Engineering
Labels can be stacked in a LSR:
– Allows an arbitrary number of labels giving possibility for multiple
control planes to act on a packet (contrast with two in ATM: VPI, VCI,
one in Frame Relay, DLCI)
Fast Recovery Possible
– by pushing and popping a backup path label on the stack in order to make
a backup LSP around the failure.
Jæger: given topic 2000
Mixing IntServ, DiffServ and MPLS?
By definition, the internet is a set of networks
connected to each other, each typically by one ISP
IntServ
MPLS
DiffServ
IPv4
Combinations, e.g.
Network Access Point
A mapping
DiffServ/MPLS
between the “QoS islands” is needed
Not standardized, work in progress
Jæger: given topic 2000
Mixing IntServ and DiffServ?
Intserv: not scalable, per flow
DiffServ scalable, aggregate
Regional/Access
IP Network
Core
IP Network
IntServ
Regional/Access
IP Network
IntServ
DiffServ
QoS Translations: Flow State <--> Relative Priority
– Service Level Agreement (SLA) which includes
Traffic Conditioning Agreement (TCA)
Edge Core Routers must apply mapping
Jæger: given topic 2000
Mixing DiffServ and MPLS?
DiffServ (modifies IP packet)
MPLS Encapsulate IP packets
DiffServ
DiffServ
MPLS-domain
MPLS domain can be seen as a link by DiffServ
MPLS can use DiffServ’s DS-field to define FEC
Jæger: given topic 2000
Mixing IntServ and MPLS?
IntServ
(Flow State) use RSVP signalling
MPLS (Labels) can use RSVP -signaling to set up
Labels and at the same time reserve resources for
the labels as in IntServ
In case LDP: translation LDP <--> RSVP needed
IntServ
IntServ
Jæger: given topic 2000
MPLS-domain
Pricing, Accounting and Charging
Previously:
– Best Effort Service
– Fixed price for Internet connection, unlimited usage
Future:
– Differentiated QoS services
– Pricing mechanism needed or else everybody will use
highest quality possibly (tragedy of the commons).
» Several schemes proposed, work in progress.
– Accounting and charging, IETF work in progress
– Management could be done by connecting to the well
established telecommunication management system
Jæger: given topic 2000
What technology to choose?
Overlay, IntServ, DiffServ, or MPLS?
Today: 80 % of the large ISP providers in US use
the overlay approach. (Although mostly for data services)
Tradeoffs exists among the technologies and type
of Internet Service Provider
– Regional ISP
– Enterprise ISP
– Backbone ISP
Jæger: given topic 2000
What technology to choose?
What is the incremental cost of doing QoS in my
network?
Consider short term / long run perspective
– Short term:
» Does statistical guarantees suffice?
» What QoS upgrades does my equipment manufacturer
offer?
– Long term:
» Some users need absolute guarantees
» Maybe change equipment provider
Jæger: given topic 2000
What technology to choose?
Regional ISP vs. Backbone ISPs
– Regional: Scaling not a major issue, can use IntServ/RSVP
– Backbone: Scaling important, aggregation needed
» DiffServ if statistical guarantees suffices
» MPLS if guaranteed service is required
Network Management functions
– Short/long term (partly vs. full management support)
– Regional/backbone (partly vs. full management support)
Jæger: given topic 2000
Summary
Technologies
for guaranteed Internet Quality of
Service (QoS) are available
Unresolved issues
– Standards for QoS translations among QoS technologies
– Standards for QoS translations among administrative
domains
– Standards for Network Management needed: Especially:
Pricing, Accounting and Charging
What
technology to choose depend upon several
factors including: time perspective, size of
network, and current manufacturer used
Jæger: given topic 2000