Re: REP writing
"Jere Retzer" <[email protected]> Mon, 24 Feb 2003 10:20:53 -0800
| Newsgroups | gmane.org.operators.rep |
|---|---|
| Message-ID | <[email protected]> |
This is a MIME message. If you are reading this text, you may want to consider changing to a mail reader or gateway that understands how to properly handle MIME multipart messages. --=_E8B78186.5131DCDE Content-Type: text/plain; charset=ISO-8859-7 Content-Transfer-Encoding: quoted-printable [email protected] 02/22/03 05:09PM wrote I put together an informal collection of thoughts (this is not an article or an essay, though I've included parts and pieces in various discussions and presentations) on regional peering in my weblog this week.=20 I generally agree with your observations, but then suppose that's to be = expected because we both promote regional exchanges. I've posted a number = of comments in response below. Apologies in advance for the non-standard = format of my responses. Our system does not support the usual forms to = mark replies. Pete wrote: >>The Internet as we know it today was built around some=20 fundamental assumptions or conditions: >>- most traffic would aggregate to relatively few centers,=20 based around population centers and computer centers, where=20 a relatively few applications would generate traffic created=20 by a relatively few major Internet businesses serving the=20 general populous=20 This is the natural Market evolution in many business services/applications= =AF the market starts at the high volume/high margin users and builds out = as it is accepted and volume expands. Dedicated data connections and video = conferencing began with high end government and large business and = migrated to the smaller. The Internet had a similar evolution but in the = case of the Internet I see at least two waves =AF the first being best = effort e-mail, web and the second is now beginning to require high = capacity, low latency/jitter for new applications including interactive = voice and video. Pete continued on assumptions: >>- most traffic ultimately ends up using a relatively few networks and = interconnect points to get from one user to another=20 This turned out to be really incorrect, especially in markets outside the = top ten such as Oregon. All you need to do is an assortment of trace = routes to see a tangled mess. If you refer to our NWAX web site, you'll = find our description of Portland routes that routinely bounce down to = Sacramento and the bay area, then up to Seattle (or even Everett, north of = Seattle) simply to go five miles across town. Last fall I tested a DSL = connection on the Oregon coast supplied by a local ILEC and found that all = their connections went to Texas before proceeding elsewhere. My favorite = was to Intel (a major employer in Portland) that went to Texas, then = Chicago and finally Seattle before finding Portland. This connection = routinely took 180 msec and dropped lots of bits. Pete continues: >>- best-effort is all we can do right now. Since the Internet cannot = ubitiqously support high-bandwidth, low-latency, low-jitter applications, = the majority, if not all, applications have to operate with the least-commo= n-denominator low-bandwidth low-reliability expectations. Applications = that require more just don't get deployed, because they don't work.=20 Classic chicken and egg problem, compounded by a lot of large-organization = inertia and denial.=20 "Not only does nobody need this but even if they did, it would work fine = =AF distance does not matter."=20 The latter statement is usually followed by the assertion that the network = is engineered to provide trans-continental connections round trip in less = than 50 msec. I disagree. The network may be "engineered" for that but the = real world does not seem to perform nearly as well, and the problem seems = to increase exponentially with the number of devices and connections = between networks. If you look at current performance on Matrix NetSystems, = for example you'll find that only a couple networks test out on average = less than 50 msec. Remember that we're also dealing with overbuilt = backbones, courtesy of the economy. Yes, I know that most of the fiber is = not lit, but I suspect if you check average utilization of the fibers that = are you will probably find the backbones operating at less than 15% = capacity. I would really appreciate current data to confirm or disprove = this.=20 In the case of the cross-country DSL connection I described above, we were = successful working through the local community getting the carrier to = connect their circuits from the community to a router that was at least in = region. This reduced average latency 100 msec. Their latency is still = excessive, but my suspected cause is local equipment at the coastal town. = I also hope to eventually persuade the carrier to do more regional = peering. The bottom line is that distance, and importantly the number and = capacity of the connections along the way really do matter. Pete continues: >>- the network build-out fizzled before the most important=20 part happened: broadband to the home. This is improving now, with=20 significant but still limited bandwidth increases through=20 DSL and cable Internet. But most of the users still access=20 the Internet through part-time low-speed dial-up=20 connections, that are all but useless for anything other=20 than email, chat and limited Web browsing. Cable and DSL are=20 usually so restricted that they offer little more than=20 always-connected faster versions of dial-up applications. I agree, but we should be wary of overstating our case. Broadband cable is = showing signs of takeoff in the local marketplace and I've seen assertions = that once it passes some magic number that people will really start = fielding applications that use it. The implicit assumption is that is when = they will fix the network. Let's hope. Pete continues >>What have we learned in the last decade, in the first=20 generation of the pervasive network? >>- everyone wants to generate traffic. This weblog is an=20 example. Peer-to-peer, distributed computing, voice-over-IP,=20 Internet radio, H.323 and SIP video conferencing, instant=20 messenging are all examples. These have all happened in a=20 network that was not designed for distributed traffic=20 generation and fought it as much as it supported it. Imagine=20 what would happen if the network supported what people want=20 to do.=20 If you research the television, Internet, and telephony markets, you will = find that the latter totally dominates from a $$ perspective. Local = telephone connections are reportedly $300b per year where the entire = Internet industry is around $30b. Entertainment is nice, but when you get = down to it real business gets done between people. Not only that most of = the interaction occurs between people who are within 100 miles of one = another. Yes, we will start doing more over networks as the networks are = able and a lot of it will be regional, in my opinion. Pete concludes: >>A regional traffic exchange is critical to these new kinds=20 of applications: >> Follows with several reasons I agree with you. Also, step back and think of the Internet from a pure = traffic engineering perspective. The simplest, most efficient network = would have the originating network pass off to the destination network at = the earliest, closest opportunity. I think this argues for regional = exchanges anywhere you can attain critical mass. Pete continues: >>- community investment. People live where they live, and are=20 invested in their communities. Texas pride is very real.=20 People do care about where their network traffic goes,=20 especially when events in far-off places affect their=20 abilities to reach their local friends and neighbors.=20 Keeping local traffic local is becoming a more powerful=20 message as people rely more on the network. I agree strongly with this statement but I'm afraid most of the US will be = slow to figure this out. Worldwide broadband deployment according to the = last report I read was over 300B but the US was only about 50B. We're way, = way behind Korea for example. I'm watching the work going on in Utah with my fingers crossed that you're = going to provide a good model. Last Fall, I wrote an opinion piece for = Peering Business News in which I proposed local FTTH and open access = regional exchanges as a solution to our current regulatory issues and also = a tremendous economic stimulus. People point out all the time, as I did = above that the backbones are overbuilt but the problem is the last mile = and the bottleneck to the last mile. I think local communities would = invest in FTTH if they new that access would be open and competitive. I don't think the FCC decision last week is going to solve the problem, = and I'm concerned that it may hurt local ISP as well as competitive = networks like Level3, XO, Verio, ELI and so on. It's all well and good to = say that we need a shakeout and that there are too many in the market but = until that happens or until we find an open access model that works no = one, including the ILECs and cable companies will make any money. It will = probably be a slow, painful process. I'm not sure how optimistic or pessimistic to feel about the FCC decision = but based upon current business models I'm pretty concerned that it is = going to have very negative impacts on the business and the economy. I see = the incumbents potentially turning off the air to the locals and competitiv= e backbones. The $$ and large traffic volumes are ultimately in connections= to homes and small business so whoever controls the last mile will = eventually own the network given the current FCC approach, or so it seems = to me. I really hope that I'm wrong and that we evolve new, effective business = models for the locals and competitive networks. My plan is to work with = them and other exchanges to see if we can help that along. I'd really = appreciate any and all ideas in this regard. Thanks, Jere --=_E8B78186.5131DCDE Content-Type: text/html; charset=ISO-8859-7 Content-Transfer-Encoding: quoted-printable Content-Description: HTML <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN"> <HTML><HEAD> <META http-equiv=3DContent-Type content=3D"text/html; charset=3Diso-8859-1"= > <META content=3D"MSHTML 5.50.4727.700" name=3DGENERATOR></HEAD> <BODY style=3D"MARGIN-TOP: 2px; FONT: 12pt Verdana; MARGIN-LEFT: 2px"> <DIV><A href=3D"mailto:[email protected]">[email protected]</A> = 02/22/03=20 05:09PM wrote</DIV> <DIV> </DIV> <DIV>I put together an informal collection of thoughts (this is<BR>not = an=20 article or an essay, though I've included parts and<BR>pieces in various=20= discussions and presentations) on regional<BR>peering in my weblog this = week.=20 </DIV> <DIV> </DIV> <DIV>I generally agree with your observations, but then suppose that's to = be=20 expected because we both promote regional exchanges. I've posted a number = of=20 comments in response below. Apologies in advance for the non-standard = format of=20 my responses. Our system does not support the usual forms to mark=20 replies.<BR><BR>Pete wrote:<BR><BR>>>The Internet as we know it = today was=20 built around some <BR>fundamental assumptions or conditions:<BR><BR>>>= ;-=20 most traffic would aggregate to relatively few centers, <BR>based = around=20 population centers and computer centers, where <BR>a relatively few = applications=20 would generate traffic created <BR>by a relatively few major Internet = businesses=20 serving the <BR>general populous </DIV> <DIV> </DIV> <DIV>This is the natural Market evolution in many business services/applica= tions=20 — the market starts at the high volume/high margin users and builds = out as it is=20 accepted and volume expands. Dedicated data connections and video=20 conferencing began with high end government and large business and = migrated to=20 the smaller. The Internet had a similar evolution but in the case of = the=20 Internet I see at least two waves — the first being best effort = e-mail, web and=20 the second is now beginning to require high capacity, low latency/jitter = for new=20 applications including interactive voice and video.</DIV> <DIV> </DIV> <DIV>Pete continued on assumptions:</DIV> <DIV><BR>>>- most traffic ultimately ends up using a relatively = few=20 networks and interconnect points to get from one user to another </DIV> <DIV> </DIV> <DIV>This turned out to be really incorrect, especially in markets outside = the=20 top ten such as Oregon. All you need to do is an assortment of trace = routes to=20 see a tangled mess. If you refer to our NWAX web site, you'll find = our=20 description of Portland routes that routinely bounce down to = Sacramento and=20 the bay area, then up to Seattle (or even Everett, north of Seattle) = simply to=20 go five miles across town. Last fall I tested a DSL connection on the = Oregon=20 coast supplied by a local ILEC and found that all their connections went = to=20 Texas before proceeding elsewhere. My favorite was to Intel (a major = employer in=20 Portland) that went to Texas, then Chicago and finally Seattle before = finding=20 Portland. This connection routinely took 180 msec and dropped lots = of=20 bits.</DIV> <DIV> </DIV> <DIV>Pete continues:</DIV> <DIV><BR>>>- best-effort is all we can do right now. Since the = Internet=20 cannot ubitiqously support high-bandwidth, low-latency, low-jitter = applications,=20 the majority, if not all, applications have to operate with the=20 least-common-denominator low-bandwidth low-reliability expectations.=20 Applications that require more just don't get deployed, because they don't = work.=20 </DIV> <DIV> </DIV> <DIV>Classic chicken and egg problem, compounded by a lot of large-organiza= tion=20 inertia and denial. </DIV> <DIV>"Not only does nobody need this but even if they did, it would work = fine —=20 distance does not matter." </DIV> <DIV> </DIV> <DIV>The latter statement is usually followed by the assertion that the = network=20 is engineered to provide trans-continental connections round trip in less = than=20 50 msec. I disagree. The network may be "engineered" for that but the real = world=20 does not seem to perform nearly as well, and the problem seems to = increase=20 exponentially with the number of devices and connections between networks. = If=20 you look at current performance on Matrix NetSystems, for example you'll = find=20 that only a couple networks test out on average less than 50 msec. = Remember that=20 we're also dealing with overbuilt backbones, courtesy of the economy. Yes, = I=20 know that most of the fiber is not lit, but I suspect if you check = average=20 utilization of the fibers that are you will probably find the backbones=20 operating at less than 15% capacity. I would really appreciate current = data to=20 confirm or disprove this. </DIV> <DIV> </DIV> <DIV>In the case of the cross-country DSL connection I described above, we = were=20 successful working through the local community getting the carrier to = connect=20 their circuits from the community to a router that was at least in region. = This=20 reduced average latency 100 msec. Their latency is still excessive, but = my=20 suspected cause is local equipment at the coastal town. I also hope to=20 eventually persuade the carrier to do more regional peering. The bottom = line is=20 that distance, and importantly the number and capacity of the connections = along=20 the way really do matter.</DIV> <DIV> </DIV> <DIV>Pete continues:</DIV> <DIV><BR>>>- the network build-out fizzled before the most = important=20 <BR>part happened: broadband to the home. This is improving now, with=20 <BR>significant but still limited bandwidth increases through <BR>DSL and = cable=20 Internet. But most of the users still access <BR>the Internet through = part-time=20 low-speed dial-up <BR>connections, that are all but useless for anything = other=20 <BR>than email, chat and limited Web browsing. Cable and DSL are <BR>usuall= y so=20 restricted that they offer little more than <BR>always-connected faster = versions=20 of dial-up applications.</DIV> <DIV> </DIV> <DIV>I agree, but we should be wary of overstating our case. Broadband = cable is=20 showing signs of takeoff in the local marketplace and I've seen assertions = that=20 once it passes some magic number that people will really start fielding=20 applications that use it. The implicit assumption is that is when they = will fix=20 the network. Let's hope.<BR><BR>Pete continues<BR><BR>>>What have = we=20 learned in the last decade, in the first <BR>generation of the pervasive=20= network?<BR><BR>>>- everyone wants to generate traffic. This weblog = is an=20 <BR>example. Peer-to-peer, distributed computing, voice-over-IP, <BR>Intern= et=20 radio, H.323 and SIP video conferencing, instant <BR>messenging are all=20 examples. These have all happened in a <BR>network that was not designed = for=20 distributed traffic <BR>generation and fought it as much as it supported = it.=20 Imagine <BR>what would happen if the network supported what people want = <BR>to=20 do. </DIV> <DIV> </DIV> <DIV>If you research the television, Internet, and telephony markets, you = will=20 find that the latter totally dominates from a $$ perspective. Local = telephone=20 connections are reportedly $300b per year where the entire Internet = industry is=20 around $30b. Entertainment is nice, but when you get down to it real = business=20 gets done between people. Not only that most of the interaction occurs = between=20 people who are within 100 miles of one another. Yes, we will start doing = more=20 over networks as the networks are able and a lot of it will be regional, = in my=20 opinion.</DIV> <DIV><BR>Pete concludes:</DIV> <DIV><BR>>>A regional traffic exchange is critical to these new = kinds=20 <BR>of applications:</DIV> <DIV> </DIV> <DIV>>> Follows with several reasons</DIV> <DIV> </DIV> <DIV>I agree with you. Also, step back and think of the Internet from a = pure=20 traffic engineering perspective. The simplest, most efficient network = would have=20 the originating network pass off to the destination network at the = earliest,=20 closest opportunity. I think this argues for regional exchanges = anywhere=20 you can attain critical mass.<BR></DIV> <DIV>Pete continues:</DIV> <DIV><BR>>>- community investment. People live where they live, and = are=20 <BR>invested in their communities. Texas pride is very real. <BR>People do = care=20 about where their network traffic goes, <BR>especially when events in = far-off=20 places affect their <BR>abilities to reach their local friends and = neighbors.=20 <BR>Keeping local traffic local is becoming a more powerful <BR>message = as=20 people rely more on the network.</DIV> <DIV> </DIV> <DIV>I agree strongly with this statement but I'm afraid most of the US = will be=20 slow to figure this out. Worldwide broadband deployment according to the = last=20 report I read was over 300B but the US was only about 50B. We're way, way = behind=20 Korea for example.</DIV> <DIV> </DIV> <DIV>I'm watching the work going on in Utah with my fingers crossed that = you're=20 going to provide a good model. Last Fall, I wrote an opinion piece for = Peering=20 Business News in which I proposed local FTTH and open access regional = exchanges=20 as a solution to our current regulatory issues and also a tremendous = economic=20 stimulus. People point out all the time, as I did above that the backbones = are=20 overbuilt but the problem is the last mile and the bottleneck to the last = mile.=20 I think local communities would invest in FTTH if they new that access = would be=20 open and competitive.</DIV> <DIV> </DIV> <DIV>I don't think the FCC decision last week is going to solve the = problem, and=20 I'm concerned that it may hurt local ISP as well as competitive networks = like=20 Level3, XO, Verio, ELI and so on. It's all well and good to say that we = need a=20 shakeout and that there are too many in the market but until that happens = or=20 until we find an open access model that works no one, including the ILECs = and=20 cable companies will make any money. It will probably be a slow, painful=20= process.</DIV> <DIV> </DIV> <DIV>I'm not sure how optimistic or pessimistic to feel about the FCC = decision=20 but based upon current business models I'm pretty concerned that it is = going to=20 have very negative impacts on the business and the economy. I see = the=20 incumbents potentially turning off the air to the locals and competitive=20= backbones. The $$ and large traffic volumes are ultimately in = connections=20 to homes and small business so whoever controls the last mile will = eventually=20 own the network given the current FCC approach, or so it seems to = me.</DIV> <DIV> </DIV> <DIV>I really hope that I'm wrong and that we evolve new, effective = business=20 models for the locals and competitive networks. My plan is to work with = them and=20 other exchanges to see if we can help that along. I'd really appreciate = any and=20 all ideas in this regard. Thanks,</DIV> <DIV><BR><BR>Jere<BR><BR></DIV></BODY></HTML> --=_E8B78186.5131DCDE--