Presentation is loading. Please wait.

Presentation is loading. Please wait.

Capacity Planning for Internet Service Networks Geoff Huston NTW Track4.

Similar presentations

Presentation on theme: "Capacity Planning for Internet Service Networks Geoff Huston NTW Track4."— Presentation transcript:

1 Capacity Planning for Internet Service Networks Geoff Huston NTW Track4

2 Issues TCP/IP Protocol Behavior Issues Usage Profile Capacity Guidelines Growth Levels

3 Planning Issues zUnderstand the domain of operation ytechnical issues ymarket issues ycompetitive issues yregulatory issues

4 TCP/IP Protocol Issues zTCP/IP is NOT a flow damped protocol yend to end flow management ysliding window protocol yadaptive flow rate designed to probe and use max available end to end bandwidth yonly limited by end system buffering size xbandwidth x delay ysystem buffers are getting larger as OS vendors come to understand the problem

5 TCP/IP Protocol Issues TCP/IP Data Flow Rate Adaptation Time Data Flow Rate Steady State Available Bandwidth Rate overflow loading into network queues

6 TCP/IP Protocol Issues zNo network-based flow control mechanism zNetwork-based packet loss signals end systems to collapse transmission window size zVarying window size allows adaptive flow metrics to adapt to changing maximum available capacity zSustained insufficient capacity leads to congestion induced collapse of data throughput

7 TCP/IP Protocol Issues zMany simultaneous TCP sessions interact with non-predictive non-uniform load ( zPeaks start to synchronize with each other zBuffering evens out individual flows, but buffers themselves behave with fluctuating load zBuffering adds latency

8 TCP/IP Protocol Issues TCP/IP efficiency under congestion load Traffic Level Data Throughput 33% 66% Congestive Collapse - The slide to misery and packet loss

9 TCP/IP Protocol Issues zTCP vs UDP zUDP-based applications yInternet Phone, Video, Workgroup zUDP Issues yno flow control mechanism ysustained use forces precedence over TCP flows yincreasing use of flow bandwidth negotiated protocols for these applications (RSVP)

10 TCP/IP Protocol Issues zDamping network capacity is not a demand management tool zNetwork capacity must be available to meet peak demand levels without congestion loss

11 Usage Profile zTwo major Internet use profiles: yBusiness use profile xpeak at xplateau yResidential dial profile xpeak at xplateau

12 Usage Profile zDistance profiles 12% Local 18% Domestic Trunk 70% International zTraffic mix due to: yDistance invisible applications without user control yDistance independent user tariff

13 Capacity Guidelines zLink Utilisation yAverage weekly traffic level set to 50% of available bandwidth. zCore network capacity should be dimensioned according to aggregate access bandwidth

14 Link Usage Profile - optimal zpeak loading less than 10% time zgreater than 50% loading for 50% time ztraffic bursting visible

15 Link Usage Profile - overloaded z90% peak loading for 45% time z60% peak loading for 60% time zno burst profile at peak loads zimbalanced traffic (import based)

16 zvisible plateau traffic load signature zsmall load increases cause widening plateau Link Usage Profile - saturated

17 Overall Growth Levels zTwo growth pressures: yserviced population ythe changing Internet service model xmore network-capable applications xusing more bandwidth

18 Overall Growth Levels zFor a constant service model the growth curve will exhibit demand saturation

19 Overall Growth Levels zFor a changing service model the saturation point will move zMore intense network use by increasingly sophisticated applications Technology shift

20 How to plan zGenerate a market demand model yforecast the number of services in operation xexisting services xgrowth rate xmarket capture level (competitive position) yforecast the average demand per service xdial access, leased line xweb, ftp, usenet xcaching trends xnew Internet services

21 How to plan zdemand models are typically very uncertain indicators yhigh level of uncertainty of externalities yhighly dynamic competitive position ypoorly understood (and changing) service demand model

22 How to plan zForward extrapolation yassume existing traffic follows a general growth model yforward extrapolat the growth model zGood for short term planning (12 months) zCannot factor ylatent demand ymarket price sensivity

23 Trend forecasting zhistorical usage vs capacity data

24 Growth Trends

25 Planning zundertake demand and trend forecast models zconstantly review the model against generated usage data zrecognise that the larger the capacity you need the longer the lead time to purchase it zrecognise that the bigger the purchase the greater the requirement for capital

Download ppt "Capacity Planning for Internet Service Networks Geoff Huston NTW Track4."

Similar presentations

Ads by Google