470 likes | 678 Views
Internet QoS : Pieces of the puzzle. 2010. Március 22. Topics:. Introduction Definitions and Terminology Internet Service Regulation QoS Architectural Issues Conclusions: Joining the Pieces Together. Sources:.
E N D
Internet QoS: Pieces of the puzzle 2010. Március 22.
Topics: • Introduction • Definitions and Terminology • Internet Service Regulation • QoS Architectural Issues • Conclusions: Joining the Pieces Together
Sources: • Internet QoS: Pieces of the Puzzle: IEEE Comm. Magazine, 2010. 01. p. 86-94 • QoS:http://en.wikipedia.org/wiki/Quality_of_service • What is QoS? http://www.tomahawkcomputers.com/qos.html • QoS Components: http://www.ciscopress.com/articles/article.asp?p=352991&seqNum=4
What is QoS? • Quality of Service • resource reservation control mechanisms • ability to provide different priority to different applications • guarantees a certain level of performance to a data flow, eg: real-time streaming • important if the network capacity is insufficient
Problem: The Internet is • unregulated • connectionless • designed without QoS! • Routers and switches cannot provide QoS cheap, fast • Best effort: default QoS level
Why QoS is not widely used? (1): • IP QoS is not Internet QoS • needed: • more architectures • scalability • IP was designed without QoS • Big growth rate • Driven by market demands • Many autonomous systems • needed: • simple • pragmatic
Why QoS is not widely used? (2): • No consensus on the exact meaning of Internet service • needed: • Understand the benefits of QoS • Good business models including QoS • Free mentality and internet neutrality • Investment in QoS cost for the end-users • needed: • flexible • rigid
QoS parameters: • Packet loss • Latency: end-to-end delay • Jitter: delay variation • Uptime: availability • Throughput: data trasfer rate
QoS categories, mechanisms: IntServ: • Reserving network resources • RSVP protocol • Not scalable, not supported DiffServ: • Marked packets by type • Queuing strategies in routers and switches • Not widely used: „the costs of premium are too high relative to the perceived benefits” BE (= Best Effort): • Not really QoS, just FIFO queuing strategy
CoS (= Class of Service): • Applying a set of priority levels • Implementation: • IntServ • DiffServ: PHB (= Per Hop Behavior): a service class • EF (= Expedicted Forwarding) • AF (= Assured Forwarding) • BE (= Best Effort)
SLA (= Service Level Agreement): • Business oriented • Technical specifications: • SLS (= Service Level Specification): guideline for impelentation • SLO (= Service Level Objetive): subset of SLS, describes the goals • Main drivers: voice and video applications • Eg: • MOS = Mean Opinion Score • ASR = Answer Seizure Ratio, percentage of calls that are successfully completed • Guaranteed uptime
Problems with SLA: • Only local agreement • Not rigorous • Presented as terms of service • without CoS • Limits the responsibility of ISPs • Little assurance of service quality
Problems with SLA – example: • 98% of • E-mail response time does not exceed 20 s • Connenction to hosted server does not exceed 1,5 s • Website download does not exceed 8 s
Types of QoS(1): • Subsription-based QoS: • Fixed bandwidth allocated • Pay for the service • Unused • On-demand QoS: • According to actual needs allocated • Pay-per-use • No guarentee
Types of QoS(2): • Soft QoS: • No service quality assured during abnormal network conditions • Not good for applications that need predictable service quality • Not attractive enough to pay • Not much better as Best Effort • Hard QoS: • Provides predictable service quality • Attractive enough to pay extra fee • Difficult for ISPs to provide QoS under abnormal network conditions
Types of QoS(3): • Explicit QoS: • The customer explicitly requires a specific service level • selling QoS as an option • Implicit QoS: • The customer does not specifically asks for QoS • Embedded into services, eg: premium service • No special fee
Other issues: • TE (= Traffic Engineering): • Performance evaluation • Optimization of networks • TM (= Traffic Management): • Network operation • TC (= Traffic Control): • Means and actions • Usage: • By vendors to implement CoS • By ISPs to offer differenciated services • By Internet backbone providers to achieve specific network performance • By IT departments to prioritize traffic
QoS planning: To be considered together: • Topology • Capacity • Traffic • Routing methods • Control schemes Aim: • To optimize network performance • To beQoS not a constraint, but an objective
Reliability: • For customers: • Availability of end-to-end functionality • For the network provider: • Ability to experience failures • No impact into the service
Regulation(1): • Internet service: • Regulate the Internet as a whole service • Regulate specific services • Eg: Internet telephony • ILECs (= Incumbent Local Exchange Carriers): basic service • ISPs: embedded service
Regulation(2): • Common standard: • Difficult • Sophisticated • Costly OAM (= Operation, Administration and Maintenance) • Regulators: • Rather penalties, than incentives • Minimum targets • Responsibility: • Service is sold by ISPs and delivered over the ILEC infrastructure • Difficult to specify responsibility
Internet neutrality vs Two-tier Internet • Internet neutrality: • Users control the content they view • good for data applications • New definition: equal treatment among similar applications • Two-tier Internet: • ISPs discriminate content according to payment • Good for delay-sensitive applications, eg: voice, video • Sophisticated • Hot debate!
Service monitoring and accounting • Monitoring: • Initiated by the operator • Obtains a general view of the network performance • Accounting: • Requested by the customer • Sophisticated • Can use the same collection tools
Internet Service Monitoring: • ISPs use self-reporting systems • Regulators can audit the ISP logs • RAQMON (= Real-time Applications QoS Monitoring Framework): to improve service quality • Internet traffic reports: • Internettrafficreport.com • www-iepm.slac.stanford.edu • Uses ping • Measures packet loss and delay • USA and developed / developing countries • 37% of the world has poor packet loss ratio
Internet Service Accounting: • No standard model • 1991: accounting framework (made by OSI) • Motivations, pros: • Provides feedback for the user • Verifies performance • Penalizes inefficient / reward efficient usage • Cons: • overhead • Detailed information about the user protection • Complex OAM, high costs
Internet Service Pricing: • Key role for success • Flat pricing: • Simple • No overhead • Low management costs but: • Over consuming • Inefficient in congestion situations • Trade-off: • ISPs want predictable incomes • Users want flexible pricing and not paying for unused services
Internet Service Marketing: Management process Identifies customers requirements Last step: advertising Trade-off: „how to meet customer requirements and satisfy in a profitable manner”
A feasible business model: Presenting value added services with the Internet service Embedded services, eg: video conference, IPTV Pricing relies on usage growth „service providers should price QoS into their services and not sell QoS explicitly”
Issues: QoS supported at application, transport, network, data link layer Some of them Where, when ,how to perform QoS routing and signaling?
DiffServ (= Differentiated Services): Scalable Efficient for streaming, file transfer Inefficient for delay-sensitive applications Factors: Cost of deployment Impact on performance Complexity of provisioning Impact on network planning and monitoring Best performance: premium traffic is a minor proportion of overall traffic
Application level QoS: Applications adjust service level Requires a signaling protocol, eg: RSVP Requires IntServ or DiffServ Problems: Scalability (IntServ) No means for negotiating the service level (DffServ) Solution: hybrid model, but not yet demonstrated
Transport Layer QoS: Compatible with DiffServe No need to involve the application Problem: a QoS level provided which is not needed or not enough TCP: Flow and congestion control: slow start, windowing Error control: retransmission, ACK mechanism Not adequate for delay-sensitive applications UDP: No delivery guarantees Requires higher level protocol, eg: RTP Adequate for delay-sensitive applications
QoS Routing(1): Assumption: BE (= Best Effort) path is used for BE and distinguished traffic No path discovery feature routing protocol needed Already existing: QOSPF (= Quality Of Service Path First) Q-BGP (= QoS-Enhanced Border Gateway Protocol) Problem: critical issues not addressed, eg: processing delay, convergence time, instability, inaccuracy
QoS Routing(2): • Solution: • scalability: SLA based on the local QoS capabilities • Single-domain QoS • MultidomainQoS • The whole Internet • Q-BGP (= QoS-Enhanced Border Gateway Protocol) • Application-layer QoS routing • Performance-based routing:best routes selected by monitoring the network performance • Multiple connections to other ISPs • AQR (= Assured Quality Routing): dynamicallyreroutes traffic
IP over QoS-driven lower layer technologies Transport technologies, eg: Ethernet OBS (= Optical Burst Switching) Traffic classification
A viable QoS Architecture Layer 7 switching: processing of layer 2 up to layer7 header Web switch: can interpret HTML tags and make decisions at layer 2 or 3 Load balancing Web caching Move content close to the end user Popular, good results
Conclusion: Customer’s willingness to pay depend on the effectiveness of the model ISP: sell QoS implicitly without special fees Adequate trade-off between penalties and incentives Goal of QoS: improve overall Internet performance QoS should rely on usage growth Selling service bundles: Best Effort should be only one bundle Not necessarily CoS Only effective when a small fraction of the traffic has to be prioritized Planning, TE, OAM, accounting, reliability play key roles
Internet QoS: Pieces of the Puzzle • Introduction • Definitions and Terminology • Internet Service Regulation • QoS Architectural Issues • Conclusions: Joining the Pieces Together Köszönöm a figyelmet!