Coming from an academic research background, I need a reality check. I would appreciate your answers/thoughts etc. to the questions I have below on WANs. 1- are there any all-IP backbones? what is the layer 2 in IP backbones (PPP, ATM, Frame relay...)? 2- how commonly is IP over ATM used? Do network operators really care about the cell-tax? 3- what is the approximate number of connections to a router (i.e., fan-in, fan-out of a router (i) at the edge, (ii) at the core, and (iii) at the backbone? 4- what is the maximum distance between any two points in a autonm. system? 5- what is the approximate ratio of copper/fiber at the edge, core, backbone links? 6- how tight is the physical space in the router rooms (i.e., is it almost a must to take out an old box in order to add one)? 7- in case of a node/link failure what is the average/approx period of time for (i) detection and (ii) recovery I thank you in advance Bulent _____________________________________________________ Bulent Yener Information Sciences Research Center Bell Laboratories, Lucent Technologies 600 Mountain Avenue, Room 2T-314 Murray Hill, NJ 07974 email: yener@research.bell-labs.com phone: (908) 582 7087 fax: (908) 582 1239 _____________________________________________________
Not necessarily. The bits would travel significantly longer distances than that assuming even nearly perfect straight line runs. If the question was in terms of equipment, (how many internal AS hops) it could be pretty bad in a non-optimal situation. (like many primary links down). Don't know whether the question as "actual, optimal", "actual, worst", "worst, worst", or "worst, actual". Deepak Jain AiNET On Thu, 14 Oct 1999, Alex P. Rudnev wrote:
4- what is the maximum distance between any two points in a autonm. system? 40,000 Km / 2 = 20,000 Km. -:)
And I know at least 1 example, surely there is dozen of such networks.
Date: Thu, 14 Oct 1999 11:02:04 -0400 (EDT) From: Bulent Yener <yener@research.bell-labs.com> Subject: IP-Internets [...] 2- how commonly is IP over ATM used?
From the ATM Forum Board of Directors Report - Rome, Italy, April 23,1999: ... Fred [Baker, chair of the IETF] stated that 75% of all Internet traffic today touches an ATM device and acknowledged the importance of the role of ATM in providing traffic engineering in today's networks. ... I don't really know where this number comes from nor whether it is really true. A number of facilities-based carriers are either migrating or talking about migrating from ATM to some sort of IP-over-fiber solution. I believe that the economics of ATM versus IP-over-fiber solutions are tremendously different for those who own fiber compared to those who don't own the fiber.
Do network operators really care about the cell-tax? [...]
Define "care". Certainly, one sees a lot of whining about the ATM "cell tax". On the other hand, if Fred Baker is correct, ATM is heavily used in the Internet. Having said that, I believe that any ISP that selects a service on the basis of some sort of overhead measurement, rather than an evaluation of cost versus performance, is confused. -tjs
From the ATM Forum Board of Directors Report - Rome, Italy, April 23,1999:
... Fred [Baker, chair of the IETF] stated that 75% of all Internet traffic today touches an ATM device and acknowledged the importance of the role
^^^^^^^ - KEY WORD. Touched - yes (this message touch ATM here twice). More interesting are the slow tendencies. The ATM as the background has the problem - the lack of QoS in this 2 level schema if don't use very complex methods (compare RED+ PRECEDENCE for the DIRECT LINK - and MLPS or ATM Bundle /cisco/). The common idea - if you have an hierarchy of the different network stacks (ATM + IP this case), and there is not good way to translate QoS markers from one level to another (in case of ATM, the only way to do it is to distribute different QoS streams by the different ATM circuits, which cause a lot of headache, and restrict the main principle KISS - don't do QoS calculations where there is not congestion). On the other hand, 2 level network structure realize another GOOD network rule - make every decision ONCE /read packet, analyze it, find outgoing device, mark it and send it there to this device - no transit points should analyze the same packets again; and it makes this 2 level schemas not so worst. The administrative issues (different administration for the background network and IP network) are the benefits (sometimes) too. Really, if someone know some deep analyze of this tendencies, please, send the reference here. Alex.
of ATM in providing traffic engineering in today's networks. ...
I don't really know where this number comes from nor whether it is really true.
A number of facilities-based carriers are either migrating or talking about migrating from ATM to some sort of IP-over-fiber solution. I believe that the economics of ATM versus IP-over-fiber solutions are tremendously different for those who own fiber compared to those who don't own the fiber.
Do network operators really care about the cell-tax? [...]
Define "care". Certainly, one sees a lot of whining about the ATM "cell tax". On the other hand, if Fred Baker is correct, ATM is heavily used in the Internet. Having said that, I believe that any ISP that selects a service on the basis of some sort of overhead measurement, rather than an evaluation of cost versus performance, is confused.
-tjs
Aleksei Roudnev, Network Operations Center, Relcom, Moscow (+7 095) 194-19-95 (Network Operations Center Hot Line),(+7 095) 230-41-41, N 13729 (pager) (+7 095) 196-72-12 (Support), (+7 095) 194-33-28 (Fax)
Bulent, I thought Lucent and Bell Labs already knew everything! :-)
1- are there any all-IP backbones? what is the layer 2 in IP backbones (PPP, ATM, Frame relay...)?
All networks which run IP are "all-IP" backbones. Perhaps this definition means "A Layer 2 forwarding infrastructure with only IP traffic". In this case, there are many. An RBOC FR network could be considered a backbone w/ partial IP, partial SNA, partial IPX, etc. Most all facilities-based ISPs run their IP networks on infrastructures with dedicated L1 TDM or WDM bandwidth. In Marketing-ese many companies, perhaps mine, say they have a "pure IP network" -- where they mean that there is no ATM or FR in the middle, no SONET APS below that, such that all of the brains and intelligence is embedded in the IP stuffs. Most of us that use to think that was a really sexy idea now think MPLS is sexy, so we're calling MPLS the 'forwarding protocol for IP' and putting some intelligence into MPLS and some into IP; so as to get the proper blend of intelligence for constructing robust, scalable, resilient IP transport networks. Except those wacky people at Qwest who just want to be different and are overtly influenced by brilliant yet obstinate swedes. All IP networks need a layer 2 framing protocol. Most use ATM, Frame_Relay, HDLC or PPP. ------------------------ = ------------------------ wr1.sfo1#show int pos0/0 POS0/0 is up, line protocol is up Hardware is Packet over SONET Description: WR2.SFO1-POS0/0-OC48 Internet address is 206.132.110.73/30 Encapsulation HDLC, crc 32, loopback not set ------------------------ = ------------------------ Some stick MPLS in the middle ala IP/MPLS/HDLC or IP/MPLS/SONET. Additionally, IP/SONET may actually be IP/DWDM wherein the Layer 2 framing protocol uses SONET frames but w/out BLSR APS.
2- how commonly is IP over ATM used? Do network operators really care about the cell-tax?
Fairly commonly, especially among 'mature established large backbones'. Approximately 5 years ago, ATM switches had OC-12 interfaces while IP routers only had DS3 or partial-rate OC-3 interfaces. I'd estimate that 40-60% of Internet traffic traverses IP networks built on top of dedicated ATM networks. This is shrinking significantly, with at least one large atm based network transitioning to IP/MPLS. Therefore the backbone trunks were done w/ ATM to provide interface capacity. Since then perhaps people/systems have invested so much in them that it's hard to move away. Another benefit to using an ATM or FR infrastructure is the Traffic Engineering (TE) ability. Network operators care about the 'cell tax' but often one can increase the overall efficacy of the network with TE to result in higher gains than are costed by the ATM 'cell tax'.
3- what is the approximate number of connections to a router (i.e., fan-in, fan-out of a router (i) at the edge, (ii) at the core, and (iii) at the backbone?
This is a function of design. In an ATM network, there will likely be 2 physical connections to the backbone ATM fabric, w/ N PVCs, in a full mesh, or truncated start topology. edge - usually 2 to the BB - 1 to 50,000 for customer aggregation varies by design core - usually 2-64 backbone links. most routers have 8-16 line interface slots. Each slot will support 1 - 4 interfaces, kinda usually... varies by design BB - see core, generally kinda the same.
4- what is the maximum distance between any two points in a autonm. system?
I suppose this is infinite. In practice, IP packets provide 8 bits for the TTL, so a maximum 'diameter' is 2^8== 256 IP hops. Note that IP hops are just that, Layer 3 opportunities for Layer 3 forwarding decisions along w/ requisite TTL decrementation. These are often hidden, in networks with IP/ATM or IP/FR, and also sometimes with IP/MPLS. This is psuedo analagous to not counting SONET switches, amplifiers, repeaters, monitors, etc...
5- what is the approximate ratio of copper/fiber at the edge, core, backbone links?
edge : 90% copper, 10 optical core: usually inter-office cabling is mostly optical, be it OC-n or gig-e. backbone: if BB ge OC-3; optical, if under OC-3, copper.
6- how tight is the physical space in the router rooms (i.e., is it almost a must to take out an old box in order to add one)?
space is one of the limited resources that an ISP has. It is generally a very significant issue. In many cases, 'silicon economics' allows folks to take out 2 boxes to put in one box that does 4 times as much. 'forklift upgrades' are not always required. utopian designs provide 'buffer floor space' such that unused rack capacity is always available for new product insertion to avoid the need for 1:1 swap-outs during real time modifications.
7- in case of a node/link failure what is the average/approx period of time for (i) detection and (ii) recovery
depends. SONET/APS is alleged to be 50ms. ATM and FR PVCs tend to re-route on the order of 1-5 seconds. IP convergence tends to re-route on the order of 30 seconds. MPLS tends to reroute today on the order of 30 seconds, but real soon now on the order of 1 second. -alan
3 bits of clarification:
Some stick MPLS in the middle ala IP/MPLS/HDLC or IP/MPLS/SONET.
This was dumb. I meant, ala IP/MPLS/HDLC or IP/MPLS/PPP. A couple of folks mailed me to state that MPLS as a transport technology was in the future, isn't it? In fact, 2 networks including our own have deployed large [mostly ubiquitous] MPLS deployments, where most traffic runs on it, one all over, another in certain regions/areas. Finally, I was 'reminded' that MPLS convergence will actually surpass SONET w/ restoral times, on the order of << 50 ms. Really real soon now. -alan
At 04:27 PM 10/14/99 -0700, Alan Hannan wrote:
Finally, I was 'reminded' that MPLS convergence will actually surpass SONET w/ restoral times, on the order of << 50 ms. Really real soon now.
Of course, the laws of physics and mathematics will continue to apply, even to MPLS. dave
That is confusing. Do the network operators actually use the mpls on Label swithcing routers (LSR) now, or the answers I am getting are for the future? Are there any LSRs with MPLS in use anywhere? I am trying to learn about the infrastructure currently deployed not the one to come in the future? _____________________________________________________ On Thu, 14 Oct 1999, Alan Hannan wrote:
3 bits of clarification:
Some stick MPLS in the middle ala IP/MPLS/HDLC or IP/MPLS/SONET.
This was dumb. I meant, ala IP/MPLS/HDLC or IP/MPLS/PPP.
A couple of folks mailed me to state that MPLS as a transport technology was in the future, isn't it? In fact, 2 networks including our own have deployed large [mostly ubiquitous] MPLS deployments, where most traffic runs on it, one all over, another in certain regions/areas.
Finally, I was 'reminded' that MPLS convergence will actually surpass SONET w/ restoral times, on the order of << 50 ms. Really real soon now.
-alan
That is confusing. Do the network operators actually use the mpls on Label swithcing routers (LSR) now, or the answers I am getting are for the future?
Are there any LSRs with MPLS in use anywhere?
Yes, a few production IP networks use MPLS on Label Switching Routers now. Probably 90% of our backbone traffic traverses an MPLS LSP. As a matter of operational concern, we do not display Layer 3 IP hops within the MPLS LSP. Therefore troubleshooting from an external perspective may be more difficult -- feel free to involve our NOC. LSRs with MPLS are in use in at least 3 networks today. 1 has an almost complete deployment, 1 has a regional deployment, and 1 uses them as 'one-offs' for traffic engineering. (as another potentially confusing detail, 'standard IP routers' are 'retrofitted' with MPLS, such that the same box becomes a hybrid IP and MPLS router/switch) -alan
Hello All, Anyone else noticing large packet drops thru either Seattle or SanJose (or both) ? Tia, JimL +-----------------------------------------------------------------+ | James W. Laferriere | System Techniques | Give me VMS | | Network Engineer | 25416 22nd So | Give me Linux | | babydr@baby-dragons.com | DesMoines WA 98198 | only on AXP | +-----------------------------------------------------------------+
participants (8)
-
Alan Hannan
-
Alex P. Rudnev
-
Bulent Yener
-
dave o'leary
-
Deepak Jain
-
Mr. James W. Laferriere
-
Randy Bush
-
Tim Salo