1. TELEFÓNICA I+D © 2007 Telefónica Investigación y Desarrollo, S.A. Unipersonal Good Things Come to Those Who (Can) Wait or how to handle delay tolerant traffic and make peace on the Internet © 2007 Telefónica Investigación y Desarrollo, S.A. Unipersonal Nikolaos Laoutaris (Telefonica Research) based on joint work with Pablo Rodriguez (Telefonica Research)

2. TELEFÓNICA I+D © 2007 Telefónica Investigación y Desarrollo, S.A. Unipersonal This paper is about shifting Delay Tolerant traffic away from the peak hour ISPs dimension their networks based on peak hour load peak hour load + QoS capital/operating cost peak hour load can be broken down to: Interactive part: web, email, YouTube, Skype Delay Tolerant (DT) part: – Existing DT: P2P high-def movie downloads, backups, OS/application updates – Future DT: Scientific datasets, e.g., from LHC@CERN DT traffic on peak hour = inefficient use of network resources Interactive traffic has a higher valuation for each unit of network resource The network treats Interactive and DT traffic indiscriminately We propose 2 ways to shift DT traffic away from the peak hour 1. Source shifting through incentives for the end-users 2. In-network shifting through a storage-enabled network architecture 2

3. TELEFÓNICA I+D © 2007 Telefónica Investigación y Desarrollo, S.A. Unipersonal Flat-rate pricing of residential broadband Currently, lots of DT traffic overlaps with peak hours P2P movie downloads Users dont bother pausing them No incentive under flat-rate pricing Ideally: DT downloads would pause during peek The movie will be seen later anyway 3 start a 3GByte download go to sleep download completes peek hour

4. TELEFÓNICA I+D © 2007 Telefónica Investigación y Desarrollo, S.A. Unipersonal Existing solutions Congestion-based pricing solves the problem But we know that users dont like complicated pricing Throttle all P2P But not all P2P is delay tolerant Requires expensive gear Publicity disasters Use capped day-time volume Requires prediction of traffic needs Assumes stationary behaviour BUT – One day you work from home – Then next you fool around with P2P 4

5. TELEFÓNICA I+D © 2007 Telefónica Investigación y Desarrollo, S.A. Unipersonal Our proposal: A flat-rate-compatible incentive scheme Flat monthly charge Ability to choose between 2 access modes on a daily basis constant : fixed maximum data rate U during the 24 hours of the day reward : U n during P peak-hours and U r during the remaining 24-P hours The incentive max daily download volume If your interactive traffic can live with U n during P and you care about P2P downloads Choose reward and youll have a much larger daily aggregate download volume 5 UnUn UrUr U rate time-of-day P flat-rate compatible non intrusive no per app scheduling efficient use of network adaptive (power of 2 choices)

6. TELEFÓNICA I+D © 2007 Telefónica Investigación y Desarrollo, S.A. Unipersonal Making the time-shift of DT traffic opaque to the end-users The previous incentive scheme requires some amount of user involvement 6 OS bandwidth cap at U n What is we dont want to involve the users? Then network has to perform the time shifting of DT traffic This requires having storage inside the network

7. TELEFÓNICA I+D © 2007 Telefónica Investigación y Desarrollo, S.A. Unipersonal Storage-enabled network architecture Large amounts of in-network storage IPOs at access ISPs TSNs at transit ISPs How it works? Users push data at full rate to their local IPO IPOs and TSNs take over SnF Two applications: A. Endogenous end-user DT data – High res images/videos – backups B. Exogenous DT data – CERN LHC data Two pricing models: 1. flat-rate 2. 95-percentile 7

8. TELEFÓNICA I+D © 2007 Telefónica Investigación y Desarrollo, S.A. Unipersonal Sending pictures to family and friends (and fooling throttling) A1: Minimize T((v,u),B,{X}) time required to send B GB of photos from v to u flat-rate subject to time of day-based throttling {X} E2E: low throughput due to throttling complementarities 8 time-of-day U available rate ε vu w X vw X wu X vw E2E SnF

9. TELEFÓNICA I+D © 2007 Telefónica Investigación y Desarrollo, S.A. Unipersonal Delay Tolerant Bulk data Is there an alternative to dedicated grids and B2: Minimize Cost((v,u),B,T,{X}) $ paid in transit for B TB of scientific data from v to u with max delay T 95-percentile with background load {X} E2E: high cost CBR pushes the background up by B/T Source scheduling can also fail due to non-coinciding load valleys 9

10. TELEFÓNICA I+D © 2007 Telefónica Investigación y Desarrollo, S.A. Unipersonal Scheduling across 2 paid links Things that affect the scheduling 10 load ISP(v) TR load TR ISP(u) C(q) q bandwidth price ISP(v) TR C(q) q bandwidth price TR ISP(u) Data: B Delay Tolerance: T

11. TELEFÓNICA I+D © 2007 Telefónica Investigación y Desarrollo, S.A. Unipersonal 11 Crossing the Atlantic from LAT to EU B=100 T=2 B=250 T=2 B=250 T=3 E2E LAT5313046 EU8149 Total6114455 SnF LAT000 EU070 Total070 Amount that the CDN has to pay ISP(LAT) and ISP(EU) (x $1000)

12. TELEFÓNICA I+D © 2007 Telefónica Investigación y Desarrollo, S.A. Unipersonal Conclusions DT @ peak hour = (economic) fat in a highly competitive ISP landscape less room for fat in our pipes flat-rate compatible incentive schemes simple way to remove DT traffic from peak hours + no change in the network - requires user attention – How capable are we in terms of ability for Delayed Gratification? SnF + storage can outperform E2E in performance for handling endogenous DTB (flat-rate pricing) in cost for handling exogenous DTB (95-percentile pricing) Therefore it can facilitate new applications (beat sneakernets?, interconnect datacenters) solve existing problems (ISP vs P2P tension) 12

13. TELEFÓNICA I+D © 2007 Telefónica Investigación y Desarrollo, S.A. Unipersonal More info at: http://research.tid.es/nikos/