session QoS vs bulk QoS Bob Briscoe Chief Researcher, BT Group Oct 2008

Transcription

1 session QoS vs bulk QoS Bob Briscoe Chief Researcher, BT Group Oct 2008

2 QoS bypass Q. bulk or session QoS? A. bulk, but BE bulk QoS I S P s End users ISPs as ASPs peak access rate vol cap DPI TCA flowspec FIFO ASPs endpoint transport resource sharing /b IP link technologies Diffserv edge AC local e2e AC policing resource sharing / scheduling QoS = differentiated congestion delay & bandwidth as link rates increase, congestion delay becoming a non-problem all the bandwidth-demanding applications are taking the QoS they need just taking a larger than average cost share of the best efforts service 2

3 the information supermarket $100/wk take what you need self check-out aisles check-outs 3

4 QoS interconnect 1,000/wk take what you need ISPs as ASPs $100/wk take what you need self check-out check-outs self check-out check-outs self check-out check-outs crosssubsidy aisles $10,000/wk take what you need aisles 10,000/wk take what you need aisles evolution by company death is too slow years need market evolution (by financial perf) months or weeks 4

5 QoS interconnection includes BE QoS QoS interconnection is not just about explicit QoS mechanisms starts with visibility of BE costs including at interconnect [Laskowski06, Briscoe05]... State of the Internet Future of the Internet 45.3% 40.0% 45.3% 40.0% 27.8% 27.8% 15.0% 20.0% 20.0% 15.7% 15.0% 20.0% 20.0% 15.7% 5.0% 7.5% 3.8% 5.0% 7.5% 3.8% Hogs Pow er Users Up & Comers Middle Children Barely Users % of Total Subscribers % of Traffic % of Cost this is how to get to this future Hogs Pow er Users Up & Comers Middle Children Barely Users Superpower Users % of Total Subscribers % of Traffic % of Cost where apps minimise cost, even if they transfer large volumes (limiting peak volume will wrongly cap BitTorrent DNA) 5

6 automatic interconnect usage cost allocation sender marks 3% of packets lightly congested link marking 0.2% of packets legend: re-ecn downstream congestion marking [%] N A marking in 2.8% of packets crossing interconnect a single flow of packets highly congested link marking 2.8% of packets N B N D receiver 6 N C

7 interconnect aggregation simple internalisation of all externalities 'routing money' legend: re-ecn downstream congestion marking [%] N A area = instantaneous downstream congestion volume bit rate solution just two counters at border, one for each direction meter monthly bulk volume of packet markings = aggregate downstream congestion volume in flows without 7measuring flows N B N C N D

8 Acceptable Use Policy Your 'congestion volume' allowance: 1GB/month (= 3kb/s continuous) This only limits the traffic you can try to transfer above the maximum the Internet can take when it is congested. Under typical conditions this will allow you to transfer about 70GB per day. If you use software that seeks out uncongested times and routes, you will be able to transfer a lot more. Your bit-rate is otherwise unlimited 2 Mb/s 0.3Mb/s 6 Mb/s bulk congestion policer operators can synthesise a carrier grade admission control service out of this see pre-congestion notification (PCN) 8 working group at the IETF differentiated services & admission control just happen as an attribute of the customer contract, not the network customer can roam without changing network 0% 0.1% Internet 0.3% congestion

9 summary everyone's got their eye on the wrong balls volume cost (congestion) AF, EF & session QoS BE QoS cost policing intra-domain & inter-domain 9

10 session QoS vs bulk QoS Q&A refs spare slides

11 more info interconnected visibility of BE cost The Internet's missing link: rest of path metrics at interconnect [Laskowski06] Paul Laskowski and John Chuang, "Network Monitors and Contracting Systems: Competition and Innovation" In: Proc. ACM SIGCOMM'06, Computer Communication Review 36 (4) pp (September, 2006) A way to do rest of path metrics [Briscoe05] Bob Briscoe, Arnaud Jacquet, Carla Di-Cairano Gilfedder, Andrea Soppera and Martin Koyabe, "Policing Congestion Response in an Inter-Network Using Re-Feedback In: Proc. ACM SIGCOMM'05, Computer Communication Review 35 (4) (September, 2005) pre-congestion notification (PCN) Diffserv s scaling problem [Reid05] Andy B. Reid, Economics and scalability of QoS solutions, BT Technology Journal, 23(2) (Apr 05) PCN interconnection for commercial and technical audiences: [Briscoe05] Bob Briscoe and Steve Rudkin, Commercial Models for IP Quality of Service Interconnect, in BTTJ Special Edition on IP Quality of Service, 23(2) (Apr 05) < IETF PCN working group documents <tools.ietf.org/wg/pcn/> in particular: [PCN] Phil Eardley (Ed), Pre-Congestion Notification Architecture, Internet Draft < (Sep 08) [re-pcn] Bob Briscoe, Emulating Border Flow Policing using Re-PCN on Bulk Data, Internet Draft < (Sep 08) these slides < 11

12 shouldn't network charge more for lower congestion? apologies for my sleight of hand actually aiming to avoid congestion impairment (loss / delay) congestion marking = congestion avoidance marking alternatively, congestion marking = price marking clearly should charge more for higher 'price marking' Diffserv example may help [Gibbens02] strict priority (g) marking probability p g q g +q b best efforts (b) weak competition price of expectation of better service 12 arbitrarily higher p g >> p b p b perfect competition price differential cost differential p g p b q b

13 Reservation enabled RSVP/PCN gateway PCN & Diffserv EF IP routers Data path processing Reserved flow processing Policing flow entry to P Meter congestion per peer Bulk pre-congestion marking P scheduled over N PCN system arrangement highlighting 2 flows SIP RSVP/RACF per flow reservation signalling table of PCN fraction per aggregate (per previous big hop) call server 1 2 expedited forwarding, PCN-capable traffic (P) (P) 3 3 (P) reserved b/w mgr 13 non-assured QoS (N)

14 PCN marking probability of PCN packets Prob 1 Pre-Congestion Notification (algorithm for PCN-marking) virtual queue (bulk token bucket) θx (θ < 1) X = configured admission control capacity for PCN traffic Yes PCN pkt? No PCN packet queue Non-PCN packet queue(s) N P Expedited Forwarding virtual queue (a conceptual queue actually a simple counter): drained somewhat slower than the rate configured for adm ctrl of PCN traffic therefore build up of virtual queue is early warning that the amount of PCN traffic is getting close to the configured capacity NB mean 14 number of packets in real PCN queue is still very small

15 value-based charges over low cost floor value-based revenue over IP, currently choice between A. good enough service with no QoS costs (e.g. VoIP) but can brown-out during peak demand or anomalies cost-based revenue B. fairly costly QoS mechanisms either admission control or generous sizing this talk: where the premium end of the market (B) is headed a new IETF technology: pre-congestion notification (PCN) service of B but mechanism cost competes with A assured bandwidth & latency + PSTN-equivalent call admission probability fail-safe fast recovery from even multiple disasters the Internet core networks could soon fully guarantee sessions without touching sessions some may forego falling session-value margins to compete on cost designed for competitive pressure towards true marginal cost app signal (SIP) per session QoS admission priority forwarding bulk data & PCN S 15 N A A N B B N D D R