Policy towards Research Networking

1. Policy towards Research Networking David Williams, CERN TERENA-NORDUnet Conference, Lund Thursday 10 June 1999 David.O.Williams@cern.ch Slides: http://nicewww.cern.ch/~davidw/public/Lund.ppt

2. Disclaimers Minefield Very personal view of how to negotiate it Few conclusions, rather initiate discussions Me, not my organisation

3. Outline • A few words on applications • Basic ideas concerning the organisational structure of networks • Some examples • Problems to fit things together • Technology evolution • Can there be any sensible policies? • Conclusions??

4. A few words on applications

5. Applications (1/2) • E-mail • Low volume (but grows with fancier attachments) • Needs worldwide connectivity (asynchronous) • General information access • Bulk of conventional Internet traffic (say 75%) • Bulk of info served from USA (mainly from CA) • Needs worldwide connectivity (weak realtime) • Data sharing and access • Important for science • Speed and reliability matter (a lot) • Reliable off-peak xfers could be scheduled • Needs worldwide connectivity (weak realtime)

6. Applications (2/2) • Remote logon • Hard to eliminate (many people hoped to…) • Needs worldwide connectivity • And human to machine performance and reliability - “as if the connection were local” • Collaborative working • Potentially vital/very valuable for organisations which depend heavily on international working • Needs worldwide connectivity • Audio and video and shared whiteboard aspects require excellent performance and reliability (tough RT constraints)

7. Basic ideas concerning the organisational structure of networks

8. What is a network - organisationally? • EITHER • a business which offers to sell network services to customers, in accordance with some more or less well-defined specifications and a price list • OR • a group of users who decide to join together to acquire network services, either by joint buying, or by assuming joint responsibility for building those services. • USER GROUPING vs SUPPLIER POWER??

9. Four main (organisational) characteristics of a net • Who can connect? • What are the goals and what traffic is allowed? • With which other nets does it directly exchange traffic? • For which other nets does it carry transit traffic?

10. Goals? Users allowed? Traffic allowed? Network A Traffic exchange directly between A and B Traffic exchange between A and C transits over B Network B Traffic exchange directly between B and C Network C

11. Who can connect? • Either anyone who pays • Or people who belong to “the user group” • In this case, do they belong • BECAUSE most of their traffic is with other group members, so network performance (RTT, bandwidth, reliability) between group members is important? • OR BECAUSE, though little of their traffic is with other group members, but they can all acquire a more cost-effective network by acting together?

12. Goals and allowed traffic? • Making a profit vs price-performance vs fair competition • Goal of ISP is to make a profit • Goal of group net is cost-effective service for members • Normally there are some (national) rules governing interactions between group nets and ISPs • Group goals vs group non-goals • Traffic profile should meet the agreed goals of group • Depending on the funding authority and the applicable rules, the profile is sometimes quite tightly defined • Major source of “problems” can come from rules for carrying transit traffic

13. With which other nets do you directly exchange traffic? (1/2) • Remember that “worldwide connectivity” is a fundamental objective of most Internet applications • So group nets typically connect to all nets with similar goals in other countries • Certainly if traffic can be exchanged for “free” • But, trans-oceanic bandwidth is (inherently) expensive, so cost-sharing is an issue • Group nets also typically connect to one or more commercial nets to handle their commercial traffic • Choosing a supplier is a price-performance decision

14. With which other nets do you directly exchange traffic? (2/2) • Commercial networks connect • “downwards” to ISPs who are willing to pay their charges • and “upwards” or “sideways” to ISPs which they can make arrangements with

15. For which other nets do you carry transit traffic? • If you are a big ISP you carry transit traffic for small ISPs as part of your normal business • If you are a big ISP you carry transit traffic for other big ISPs that you have reached arrangements with • If you are a group net you typically carry transit traffic when you think that the arrangement will be mutually beneficial

16. Some practical examples NRNs amd TEN-155 (US) Agency networks Internet-2 / Abilene vBNS Commercial networks Specialised networks

17. National academic & research nets (1/2) • Who can belong? • Decided on a case-by-case basis in each country • Always for research universities • Nearly always for research labs (but not in US) • What about non-research universities? • What about secondary schools? • What about primary schools? • What about libraries? • What about local government offices? • What about commercial companies “spun-off” from universities or research centres?

18. National academic & research nets (2/2) • Goals and allowed traffic • Good service for group • So, good site-to-site performance is important • General (commodity) information access important too • Traffic exchange? • Other nation’s A&R and related nets (agency nets, TEN-155, etc) • One or more commercial ISPs • Transit traffic? • Not much

19. TEN-155 • Who can belong? • European NRNs + …. • Goals and allowed traffic • Exchange among members according to AUP • US traffic handling for some members • (MBS) • Traffic exchange? • See below • Transit traffic? • Carries traffic between NRNs and non-European (Japan, USA?, …) nets handling A&R-related traffic, where agreements are in place

20. Agency networks (1/2) • Background • In USA several agencies are funded by federal government • And universities are either private or funded by state governments • Strong funding separation • Who can belong? • Agency funded labs and projects • Goals and allowed traffic • Defined by agency mission

21. Agency networks (2/2) • Traffic exchange? • To get to sites important for agency mission • Transit traffic? • Hardly ever (I think), on the grounds that its not their job to carry that traffic • One example is that ESnet does not carry transit traffic from (European) NRNs to US universities

22. Internet-2 / Abilene (1/2) • Who can belong? • UCAID members (for the Internet-2 aspect) • Who have also paid up to use Abilene (maybe with help of NSF grant) • Goals and allowed traffic • High performance connectivity between US research universities • So, in principle, no “commodity” traffic

23. Internet-2 / Abilene (2/2) • Traffic exchange? • With vBNS and US agency nets • With international NRNs • According to local (not Internet-2) decision, with any ISP or net present at the GigaPoP • Transit traffic? • Only according to I-2 exchange agreements as discussed in previous point

24. vBNS (1/2) • Who can belong? • NSF grant holders • Who will normally (not necessarily) be UCAID members • Goals and allowed traffic • Support of NSF collaborative projects requiring high performance connectivity • In principle, no traffic unrelated to those projects • A rather tight AUP

25. vBNS (2/2) • Traffic exchange? • With Abilene and agency nets • With international NRNs connected via STAR-TAP • According to local decision, with any ISP or net present at the GigaPoP • Transit traffic? • Only according to I-2 exchange agreements as discussed in previous point

26. Commercial networks • Who can belong? • Anyone who pays • Goals and allowed traffic • To make a profit • No (few) restrictions on traffic • Rather strong flow to/from US • Traffic exchange? • Decided on commercial grounds • Transit traffic? • Decided on commercial grounds

27. Specialised networks, such as ANX • Who can belong? • Members of the club • Goals and allowed traffic • Run a highly performant Internet for (US?) automobile companies and their suppliers • By qualifying ISPs and exchanges, and monitoring their performance • Traffic exchange? • Gateway to public Internet • Transit traffic? • No

28. Problems to fit things together Congestion Packet loss Exchange points

29. Daily packet loss structure on a congested route 8 loss-free hours at night From ~01.00 to 09.00 CET 50% peaks Thurs Tue Fri Mon Wed Sat Sun Sun

30. FNAL-Brown connectivity - fixed in June/July ‘98

32. (Some) things do improve with time!!

33. Catastrophic overload • Getting QoS/DS into production will be a very welcome advance • But I suspect that no amount of QoS/DS will help on links which are catastrophically overloaded • Roads, cars, railways, phones are typically in use for less than 10% of the time, or deliver only some 10% of their theoretical bandwidth, when measured on a 168 hours/week, 52 weeks/year basis

34. Exchange points • The government structures are very different in Europe and in the USA. That leads to network goals and rules for allowed traffic which don’t match well • Deciding at which points to exchange traffic is not so obvious • And the cost of getting there may be (very) high • Leads to difficult discussions on cost sharing

35. Technology evolution Again, personal views

36. Since last year in Dresden (1/3) • Quick move to TEN-155 in Europe • Most national nets in Europe deploy or plan 155 Mbps or faster backbones • Internet-2 has become a reality with vBNS (622 Mbps, soon to move to 2.5 Gbps) • I-2 starts to deploy a multiple backbone architecture - Abilene (2.5 Gbps) plus vBNS, using GigaPoP as a basic component • QoS (still) coming…..

37. Since last year in Dresden (2/3) • Deregulation in Europe reduced some prices quite quickly • N. Atlantic pricing also fell a lot • Concept of an (all) Optical Internet became concrete, most obviously in Canarie-3 • Based on HDWDM technology, headed towards 100s or 1000s of s • May completely change how you build MANs (80-100 km without any amplification) • Gigabit Ethernet is winning many commercial battles as the line protocol

38. Since last year in Dresden (3/3) • But ATM (VCs) more widely deployed as backbone technology and that helped quite a bit X-At by reducing hop counts • And in some critical Internet Exchanges • Initial MONARC work convinces me that we are working in a very challenging area of distributed computing with really huge data volumes where few companies are active

40. Can there be any sensible policies? Worry more about Interconnects Worry less about commodity traffic Improve our technical understanding Concentrate on services for A&R

41. Interconnects • Because of poor competitive situation we have tended to worry mainly about line costs in Europe • We should worry at least as much about the places where we interconnect • All A&R sites in an (extended) city should work together to create a GigaPoP

42. GigaPoPs • Where you separate out your traffic (and if that’s hard, let’s work on it) • And send different types of traffic to different nets (including ISPs) • And encourage competitive presence of suppliers of bandwidth • Including competitive suppliers of s and of (dark) fibre

43. Commodity traffic (1/2) • How wise is it that European NRNs carry long-distance commodity traffic? • 10 years ago, this was essential, since ISPs could not provide the service (at a sensible price) • My inclination tells me NRNs should try to offload commodity traffic as quickly as possible • Provided it can be done at reasonable price-performance • NRNs for bulk-buying compared to NRNs as ISPs for their local community

44. Commodity traffic (2/2) • They would still handle all of the traffic for NRN customers, just hand commercial part off to an ISP offering the best price-performance service as soon as feasible • Otherwise the distinction between NRNs and ISPs gets very blurred • And the NRNs are vital for maintaining a sensible A&R network inside each country

45. Wider understanding of Internet traffic • Instrumentation of the Internet and understanding of its performance has vastly improved in the past ~18 months • Know a lot about RTT and packet loss rates on many routes • Know much less about the integral traffic flows and the engineering of Interconnects • Work to do there

46. Services for A&R • Another global trend is that ISPs start to offer many (optional) services to their customers. • Such as e-mail hosting, Web hosting, home directory support (backup etc), …, and I expect this trend to grow. Internet Service Provider ==> Application SP; ISP to ASP. • NRNs natural customers are a quite well-defined population with some clear application service needs. • Should make sure that they are well taken care of

47. Conclusions?? Just defined what I regard as sensible policies What about the European dimension?

48. The European dimension (1/2) • There are very important roles for the NRNs, DANTE, TERENA and the EU to all play in not losing the momentum created by the success of TEN-155 • If each country had a set of well-run GigaPoPs • Which were interconnected nationally • It would help enormously in creating a competitive situation for interconnecting them on a pan-European basis

49. The European dimension (2/2) • The EU needs to be especially careful of thinking that you can ignore the A&R community and build an e-commerce infrastructure in glorious isolation • Funding 2-3 labs to study Internet protocols and performance in the real world would be really very useful • As would getting to understand how much Europe could benefit from a very broad collaboration between suppliers, industry, commerce, academia and research labs