Context n The session is about Research Network Policies n Hence not about Global Internet Policy (how society should try to deal with the Internet Age) n Nor about Research Network Practice, except to illustrate ideas n My title is Do policies Help more than they Hinder? n A little provocative perhaps, but I want to make us all try to think a bit
Networking is EASY n We all know what we want…. n And can easily explain the details…. n We want networking which is:- –Fast –Reliable –Easy to use –Offers lots of facilities, including video and audio streams, searching, easy info publishing, and video-conferencing –Connects to all the people and organisations around the world that we need to work with (MicroSoft, RedHat, Sun, Oracle, all our collaboration, including the sites in Pakistan, China and Tashkent, our favourite magazines and papers, our friends …) –and is CHEAP
Networking is HARD n End-user performance and reliability typically depend on multiple different organisations n Not all of which are interested in solving your (users) problems (except maybe at a high price) n High growth rates mean that good performance today is no guarantee of good performance in one months time. n If you look after 1k users they will each have at least 20 different destinations that they use a lot, and there will be little correlation among those. So you have to deal with maybe 10k key routes. n For 100k or 1M or 10M users its a lot worse n Frightening speed of technical change n True 24x7 operation
POLICIES - HELP OR HINDRANCE? Some general ideas from IT, not from networking
When can policies help? n When they encourage the formation of consensus –Such as enterprise or discipline-wide decisions to define supported platforms –In 1980s until early 1990s many companies had a policy that they tried to do (almost) all of their computing on VAX-VMS –Now more likely to be pushing Linux and/or Open Source –Or specific design packages n Or when they really save money –We should have a natural suspicion of claims that policies save money, since it is very hard to measure the total cost of ownership –The best thing is to try to encourage competition n Subsidiary issue –How do you police, encourage, implement…. your policy?
And when can they hinder? n If they take you down a blind alley n Or if they discourage effective competition n Examples of hindrance –Mid-1980s Buy OSI networking solutions policies. Well intentioned but very expensive, since it was a blind alley. –NatWest Bank buying into MSs pitch to build your enterprise computing solution –Government policies in favour of their local telecoms monopoly n Of course, people change their policies with time, and as the technology situation evolves
THE LIGHT AT THE END OF THE TUNNEL? DENSE WAVE DIVISION MULTIPLEXING
Why could one technology have such a major impact?? n Fibre as a waveguide, with well behaved photons - your own private and sheltered aether n The discovery that you can actually deploy and operate multiple wavelengths in the real world n A dynamic new industry, growing fast and attracting lots of capital investment n Can rapidly deploy huge bandwidth upgrades in suitable existing pipes n People really thinking that we can very soon transmit 40 times 40 Gbps on one fibre n We came a long way from 64 kbps connections in 1985, or even from expensive 2 Mbps in 1995
So what changes? n Economics of providing bandwidth –Estimated that progress in photonics is 2x per year, compared to 2x per 18 months for electronics Adds up over time! n The balance between packet switching and circuit switching –You can think of creating economic and flexible paths between source and destination –Fully optical switching of IP packets may be a long way off –And converting between optical and electrical regimes is expensive n The old default assumption that someone else (a telecoms carrier) worries about transporting your traffic –No regulatory reason why optical transport should be a quasi- monopoly –Just becomes another advanced service that our community needs to worry about acquiring, directly or indirectly, via (long-term) lease or other suitable direct or indirect deal –Connecting one site to its 10 major peers becomes feasible
ARCHITECTURAL ISSUES INTERCONNECTION POINTS and TRAFFIC FLOWS
(Inter) Connecting Regional MAN SiteWide LAN National A&R Service International A&R Interconnect Commercial ISP - International Service Disciplinary Network 1
Architecting the (inter)Connections Regional MAN SiteWide LAN National A&R Service International A&R Interconnect Commercial ISP - International Service Disciplinary Network 1 Ipoint1 IPoint3 IPoint4 IPoint2 How is this handled? Ipoints 1-3 may well be co-located and/or there may not be a Regional MAN
The structure of commercial networks n I would be interested to understand the extent to which the commercial networks operating in Europe do or do not have a national substructure. Does the interconnect structure conform to the traditional national borders, or does it ignore them. [Must be some lessons from looking inside the regions of D, F, UK,…..] n If it is potentially more efficient (as I suspect) to ignore national boundaries, is it wise to cast them in concrete inside our NRENs? n Point for discussion?
Where should we split the traffic? n Should an A&R site immediately hand-off traffic destined for commercial ISPs? Or leave it to the NREN to do that? n Many people feel that it must be a good idea to split traffic early, to optimise the crucial A&R traffic performance n But it is work for the onsite network staff (may be few and/or part-time and/or none) n So is it better to hand all traffic to one (or few) suppliers who take care of everything n Point for discussion? n I am not convinced that the IP community worries enough about identifying traffic flows by type
SOME POLICY ISSUES Starting with some basic questions and my personal answers Starting with some basic questions and my personal answers
We can ask n Is A&R networking inherently different from commercial networking? Why?? n If there is a distinction today, will it stay like that for the next five years? Always?? n How similar, or different, is A&R networking from A&R computing?
Some personal answers (1/3) n Is A&R networking inherently different from commercial networking? Why? n Traditionally we have said –We are interested to explore our advanced needs, and are able to prototype services which are less than perfect –Commercial ISPs have failed to demonstrate good understanding of A&R needs, and have failed to provide good service to the A&R community when given the task –Academia and research have much higher needs than the home and commerce n The first point is true, the second has been true and might change with time, and I am not so very convinced about the third n BSA has already provided a very good positive answer to the question
Some personal answers (2/3) n If there is a distinction today, will it stay like that for the next five years? Always?? n Who can tell? Unless a vendor sets out to just tackle the A&R market (seems unlikely to me) the most plausible scenario is that companies become very successful in the very advanced commercial market [lots of VC, streaming and RT audio, multiple sites all over the world, lots of connections to individuals] and then get interested in the A&R market. How similar are those markets?? n Again, BSA tells you that you should be helping the local community, and that the big suppliers just dont get it.
Some personal answers (3/3) n How similar, or different, is A&R networking from A&R computing? n My experience tells me that A&R computing (scientific computing) was very different from commercial computing 30-15 years ago, and has gradually converged with commercial computing n By and large the important companies understood the A&R computing market. Not much sign of the same thing in networking yet. n But one huge difference is that a single site (university, research centre) could largely run an independent computing policy while it has to cooperate with its peers on networking policy
SPECIFIC THOUGHTS ON POLICY FOR EUROPEAN A&R NETWORKS
In Europe we need - Competition n One of the potential pitfalls of policy is that it can sometimes have the effect of inhibiting competition n It would be good if A&R sites and NRENs concentrated more of their thinking and resources on their Interconnection Point(s) n Should be well-engineered, but neutral (favouring neither a telecom vendor nor an ISP) n Offering good connections to the regional MAN n And to the national A&R service n And, in general, be flexible in meeting the site or NRENs interconnection requirements n Full competition in provision of connections (links) n Potentially competition in provision of the IPoint too
In Europe we also need … n Access at a sensible price to one (or, better, two) fibre pairs passing all European capitals n A good way of building technical and organisational consensus about the architecture of the interconnects and how to fund and manage them n And a agreed way to evolve that consensus as the technology and market changes n (more follows)
We should also… n Strongly develop our neutral interconnects and to encourage many service providers to connect to them n Concentrate more and collaborate better on applications n A huge potential here –Grids –A&R Portals –Collaborative tools –…..you name it…. n After all networking is only a tool at the service of another goal (science, education, society, individual people, ….)
You have to be optimistic n And hope that we can reach consensus on sensible policies which will HELP n Counter examples have existed. Lets try to avoid them in future. n Keeping to simple, pragmatic concepts tends to be useful n Try to encourage competition n And encourage local infrastructure builders n Explain your requirements, problems and hopes to any service suppliers and local, regional or national governments that are smart enough to ask n Creating the A&R Internet we would like is still a lot about scientists getting involved in politics and society, and influencing commerce and industry