Differentiated Services - Panacea of Networking?

Transcription

1 Differentiated Services - Panacea of Networking? Kalevi Kilkki Principal Scientist, Nokia Docent, HUT NOKIA DiffServ / December 1999 / KKi page: 1/34

2 Panacea pan a ce a (p²n -s ) n. A remedy for all diseases, evils, or difficulties; a cure-all. [Latin panac a, from Greek panakeia, from panak s, all-healing] Diseases to be cured overall weakness of end-to-end services Evils to be tamed those who try to exploit network resources in unfair manner Difficulties to be overcome implementation and management of complex networks NOKIA DiffServ / December 1999 / KKi page: 2/34

3 Outline Identification of of problems Identification of of solutions Non-IP solutions IP IP evolution based on on hard guarantees IP IP evolution based on on soft differentiation NOKIA DiffServ / December 1999 / KKi page: 3/34

4 1. Identification of problems 2. Identification of solutions Non-IP solutions 3. IP evolution based on hard guarantees 4. IP evolution based on soft differentiation NOKIA DiffServ / December 1999 / KKi page: 4/34

5 Service demand Problem: Demand S10 S19 Intrinsic Problem unpredictable and highly variable service demand (time & destination) S1 destination time NOKIA DiffServ / December 1999 / KKi page: 5/34

6 Question: QoS vs.. Demand Should QoS be independent of service demand? Telephone network YES, quality is, by and large, independent of demand too many customers at the same time in the same place some customers get service while others get nothing Internet NO, quality depends essentially on demand too many customers at the same time in the same place everyone gets poor service NOKIA DiffServ / December 1999 / KKi page: 6/34

7 QoS vs.. Demand vs.. Price Quality (technology)? Demand (behavior) Price (business) NOKIA DiffServ / December 1999 / KKi page: 7/34

8 Service demand time 7 13 Operator s choices 19 S1 S10 S19 destination "Throw bandwidth" approach: Put enough capacity to to make everyone content Business approach: Try to to control the demand by by price Technology approach: Share what is is available as as fairly as as possible NOKIA DiffServ / December 1999 / KKi page: 8/34

9 1. Identification of problems 2. Identification of solutions Non-IP solutions 3. IP evolution based on hard guarantees 4. IP evolution based on soft differentiation NOKIA DiffServ / December 1999 / KKi page: 9/34

10 2 Networks 2 Service Models Guaranteed Best Effort Circuits Telephone network? (maybe (maybe possible possible with with CDMA) CDMA) Packets ATM IntServ + RSVP Internet Differentiated Services? NOKIA DiffServ / December 1999 / KKi page: 10/34

11 Best Effort & Guarantees - 3 Ways Two separate systems On the basis of circuit model On the basis of packet model NOKIA DiffServ / December 1999 / KKi page: 11/34

12 1. Identification of problems 2. Identification of solutions Non-IP solutions 3. IP evolution based on hard guarantees 4. IP evolution based on soft differentiation NOKIA DiffServ / December 1999 / KKi page: 12/34

13 Start 1: 2 separate systems Each application uses circuit or packet network seems to work - so we should not omit this approach totally! however, some applications need features not provided by one network no efficient, dynamic multiplexing double trouble (services, interfaces, management, billing) NOKIA DiffServ / December 1999 / KKi page: 13/34

14 Outcome 1: Pressure to integrate Let us design a new network (A) to avoid the shortcomings of old ones that was the idea of ATM The risk instead of integrating two networks a third one is created, and then all three try to live with each other even more interoperability problems and overheads Another alternative take an existing network as a starting point and improve and expand it T T A I I NOKIA DiffServ / December 1999 / KKi page: 14/34

15 Conclusion 1: Instead of designing a new network, an evolutionary approach based on packet network (IP) is is the only reasonable choice NOKIA DiffServ / December 1999 / KKi page: 15/34

16 1. Identification of problems 2. Identification of solutions Non-IP solutions IP evolution based on hard guarantees IP evolution based on hard guarantees 4. IP evolution based on soft differentiation NOKIA DiffServ / December 1999 / KKi page: 16/34

17 Start 2: Based on the current Internet model Best Effort Service far from perfect, but seems to work well enough for most purposes continues to make the fundamental Internet service Features no quality differentiation flat rate charging no reservations for flows simple traffic control in core network relies on TCP/IP no hard control over traffic process As many packet as possible are delivered Packets are dropped only if necessary NOKIA DiffServ / December 1999 / KKi page: 17/34

18 Outcome 2: Pressure to add Guaranteed Services Approach: In addition to best effort service, add guaranteed pipes Pipe for the flow through the network Perfect Quality of Service (for accepted traffic) Might be dropped whenever needed (and even if not necessary) NOKIA DiffServ / December 1999 / KKi page: 18/34

19 Integrated Services & RSVP - a short analysis Guaranteed, Controlled Load & RSVP with exactly defined bit rate and strict QoS objectives (like ATM) capacity reservation is required throughout the network charging is based on the bit rate and QoS objectives NOT a small evolutionary step instead, it means a huge change of service philosophy Acceptable only if we can find really good reasons (1) for resource reservations (2) for complex charging NOKIA DiffServ / December 1999 / KKi page: 19/34

20 Reservations - What is the point? IP is a packet networking technology capacity reservation is not a natural part of IP credible reasons are needed to justify reservations Typical statements customers are willing to pay extra only if something is reserved Wrong! Reservation is a mechanism rather than a service improved QoS is not possible without capacity reservations Wrong! Sometimes capacity reservations are harmful reservation sharing NOKIA DiffServ / December 1999 / KKi page: 20/34

21 Applications needing reservations Priority for flows in progress traditional telephone network is the most obvious example reasonable objective with some applications (audio, video), while not relevant or even undesirable with some others not the default operation in IP instead, higher priority can be used permanently for important flows Virtual Private Networks (VPNs) reservations make it easier to design and control Service Level Agreements from business viewpoint reservations seem useful but bring about low utilization (<10 %) NOKIA DiffServ / December 1999 / KKi page: 21/34

22 Complex pricing - Why not? A lot of small flows Web-browsing short duration per destination IP telephony small bit rate $ 0.03 $ 0.13 actual transmission of bits is not usually the dominant cost factor complex because parameters related to QoS are needed as well $ 0.03 Average bit rate = 124 kbps, number of packets = 115, number of bytes = 74 k, lost packets = 1 / 9, average delay = 78 ms, maximum delay = 111 ms destination = , required quality = A3 billing & accounting system yields large expense NOKIA DiffServ / December 1999 / KKi page: 22/34

23 Conclusion 2: Guaranteed Services may have minor role in in future IP but surely not a major one NOKIA DiffServ / December 1999 / KKi page: 23/34

24 1. Identification of problems 2. Identification of solutions Non-IP solutions 3. IP evolution based on hard guarantees 4. IP evolution based on soft differentiation NOKIA DiffServ / December 1999 / KKi page: 24/34

25 Start 3: Service differentiation Differentiated Services Panacea - perhaps but then some problems should be solved Overall weakness of end-to-end services best effort is not enough for all applications differentiation is needed for business reasons Exploitation of network resources TCP/IP model is not always robust and fair Implementation and management problems should be solved but without any excessive implementation and management costs NOKIA DiffServ / December 1999 / KKi page: 25/34

26 Quality for IP - An implementation IP is a packet network let us avoid converting it to a circuit switched network A packet can be damaged, delayed or lost to manage packet losses importance scale for packets (I) to manage delays urgency scale for packets (U) bit errors are relevant sometimes but not usually in core network IP network time NOKIA DiffServ / December 1999 / KKi page: 26/34

27 Edge functions Task is to define the basic attributes for each packet not for flows because IP/DiffServ core is a packet transport tool Basic attributes importance of the packet (I) urgency of the packet (U) All needs are mapped to these attributes higher price more important more demanding application more important (but this is not exactly the standard DiffServ model) NOKIA DiffServ / December 1999 / KKi page: 27/34

28 The essence: Edge model Business Business models models Management Marking system A I=x,U=y to next node I,U F(I,U) Marking system D I=z,U=v edge functions core functions NOKIA DiffServ / December 1999 / KKi page: 28/34

29 Simplest part: Core functions Load situation Queuing + Difficulties Difficulties to to be be overcome overcome I,U I,U Discarding decision F(I) F(U) NOKIA DiffServ / December 1999 / KKi page: 29/34

30 Service differentiation Several levels of fixed price, e.g. $100 per month $30 per month $10 per month Relationships between bit rate quality (delay) availability Bit rate 3 M 300 k 30 k low Quality Diseases Diseases to to be be cured cured high NOKIA DiffServ / December 1999 / KKi page: 30/34

31 throughput Quality vs.. Demand - incentive to adjust bit rate Optimal region for the user Evils Evils to to be be tamed tamed Idle route or time Moderately busy route or time Busy route or time Traffic sent by a user (fixed price) NOKIA DiffServ / December 1999 / KKi page: 31/34

32 DiffServ - Panacea? Diseases to be cured: weak services Yes, DiffServ provides basic tools for service differentiation Evils to be tamed: exploitation of resources Yes, if the packet marking is properly designed Difficulties to be overcome: complexity Yes, as to the core, while edge functions and management are unsure NOKIA DiffServ / December 1999 / KKi page: 32/34

33 References More (356 p.) about DiffServ M TP MACMILLAN TECHNOLOGY SERIES DIFFERENTIATED SERVICES FOR THE INTERNET Kalevi Kilkki ISBN NOKIA DiffServ / December 1999 / KKi page: 33/34

34 NOKIA DiffServ / December 1999 / KKi page: 34/34 The End