Live Transcription - Tuesday - 9:00 - 10:30
Note: Please be advised that this transcript is the output of the real-time captioning that was used on a trial basis during the RIPE 55 Meeting. In some cases it may be incomplete or inaccurate due to inaudible passages or transcription errors. It is posted as an aid to understanding the proceedings at the session, but it should not be treated as an authoritative record.
THE CONFERENCE COMMENCED ON THE 23RD OF OCTOBER, 2007, AS FOLLOWS:
MR. HUGHES: I would like to introduce our first speaker today, real world stuff, and Fernando Garcia is going to talk to us about traffic engineering in a multi-homing environment and if you press the button the little switch box you should go to the thing there. Here we go. So Fernando, thanks.
SPEAKER: Good morning. OK, this title is really a but what is really this presentation about? This presentation starts is based on a real solution we make for our real customer, and is equally running in a real scenario that I will show you later. Of course, the name of the is for me to a voice product. Has to realise call them A and B, and he may be coming in because he want to but he wanted to get optimal uses for the lines. OP mat for most of the getting well lines the other have. Many times this is well as if you go to some fizz I can't they tell you that you must follow what the traffic says, the traffic says this line is full increase the capacity of this line, see if the other line gets full increase the line of the other capacity. But many times we see that the customer has different opinions. Many times it's practical, OK I have some agreement with this carrier so this must be my main carrier, so please, do the traffic, go with this carrier. This scenario is different because this customer is basing the DFSH nowhere and it was impossible for him to get anything bigger than a three excuse, than three lines from any carrier, so he really needs to distribute the traffic in a way with both carriers.
In any case, I of course, as a service provider I needed to follow the customer requirements, you know the customer always is right, but I always try to be a good citizen, that is don't aggregate, don't do all the things that we really hate, so let's try as the customer let us play with his networks well, he let me play in off peak hours but he let me play. This is good because it is not base I didn't base my ex TERMENTings in a laboratory but using his customer and his traffic. Before I start, this is the traffic he received in both through both carriers and you can see carrier B is in peak hours, and they like yes, they liked, there is carrier base always good, is reaching the maximum, well we can call these this is when when I make the test, this is called mainly . But the way this is 28, so this carrier, didn't reach the maximum.
So I think you know these but in case you have forgot because it's a long, long time, how the BGP select the traffic base in all these parameters but the only two that really get distributed through all the Internet, yes we can call it, are the shorter AS path and the lowest origin type. So I try to make some experiments using these ones and things that I will tell you now.
And one thing I want to tell you is very easy: If you are content provide certificate very to balance usually you can choose how your traffic goes out of your network but it's more difficult to select how the traffic goes into your network. In this case, my customer was not a content provider but a provider for seating, so he has thousands of customers that contends read mail and use and the traffic in this case, the traffic in common is more difficult to influence. Sometimes you can and in this case I made it, you can choose the same method for both input and output, sometimes you must use different methods or different things in input and in output, but I will concentrate on input traffic that is usually more difficult.
How I measure and reach the final solution: Well this is the initial solution. The customer has unibare router. I made in fact I make clear recover it, and then I measure the traffic through both lines and then I make exactly the same thing through the other line. You say why? It should be the same. Well I thought that but I discovered that it's not how to do that way. In the first step, this is what I get. You see that 45 for the traffic but after the second cut this is what I get, 29 of the traffic. The difference, about 16 percent is what we call, saith traffic that is set don't matter. It can go both ways so I will reach the more stable one and this is a problem. Well in the end after two or three weeks, one or two one, two weeks all the traffic gets this way but it takes more than a week to reach this point. So it's traffic that must clearly say that both parameters, both roles have the same weight. I prefer the more stable one and so it's difficult to use some in this situation. You can see this is table traffic because if you have any good you can reach this and this is very bad because the traffic gets separated.
So I make the solution that we always try, at least I always try because it's the one that they tell in how the BOOS is pre PENDZ. You add your prefix and SEFRLS times. One, two, I have seen eight times. This is what I mention in router, this is of course Juniper, you know how to do it I am sure and what I discovered is that you have too much influence. You see, this is the initial situation and when I make suddenly it went all the other way. No, I don't want this because then we have the problems in the other line. And we see we still have here, about 7 percent of don't matter traffic. So I said OK let's try other things. Deaggregate, oh, yes, I know. In this case, I have the luck that this must have a very big style base so he can make in fact 3 requests KSH for he has three not 3 separate prefix and he needs to denounce the three so I could play with the three prefix differently and still be a good citizen. So I announce in each part with different prop ends or attributes and it's complex to apply and not in many cases, you only have one prefix it's not good net Zen Vision ship because you are deaggregating and poly filing the role table.
XXX
This is one example of what I meant. Of course the [ ]are not those really and this is what I get. I run out of [ ] I reach not so extreme results but in any case, too much, and of course we still get the 9.9 percent don't care traffic. Then when I said OK, third try, I will use communities, oh but because one of the carriers didn't have communities, didn't have communes, the other one has communities and WB I try to Tuesday I discovered that they didn't work so I couldn't try and that is the reason I don't have graphics to show you.
Then I try something that is not very used, at least I haven't it too much but I said well let's try, it's to use origin. You know, its route can have origin internal, external or incomplete. Usually it's set by the router and you don't mind, you don't care what it's set. OK well I want to play with it. This when AS path are equal, this happened frequently, in fact I have seen it happen much more frequently than I will think, so a lot I don't know if he is questioning with my customers or with his but I have seen lot of situations that you get or do announce your prefix and many places they get the so I said let's try if we can play with the origin, and we play with the origin it's important because we can influence in the don't care IS so we said announce one year to internal, less traffic incomplete with the other. As you know, as you must know INTESHLly get preference over incomplete. And what I get is this: This is the original, this is the last one, we see that we have a more balance traffic distribution and, very important, we don't have we don't have almost any, almost nothing of I don't care traffic. The traffic is very stable, when the same, when you cut one way or the other, so it's a very good solution. What I may finally, well for I said I have to show you, but the customer said we are still get too much traffic on this carrier, I said OK, I apply origin selectively to some prefixes and to others and these as I told you, not aggregated traffic but prefixes assigned by RIPE so I still keep being a good net citizen. And finally I get this traffic: I can it's very good, and we are almost don't get well there are some little here but not too much and both lines are almost full 100 percent in peak hours.
What I will say to you, well, as a gene solution, of course KSH the gross add adjust can be made with pre-penned but you have seen this very, very influence so to fine tuning you can set the origin and to make a final tuning, selectively apply, pre-penned and/or origin to some of the prefixes and not to the other ones, so one TIENG and one prefix, so this is mainly the presentation I was made with. To give you some ideas if you have some questions, I didn't get any signal that I was finishing later so.
XXX
MR. HUGHES: Anybody got any questions? Yes, a question at the back right. [Chris] tan RIKS en. Many of the techniques that you have described are best stable configurations for traffic but then on the other hand, user traffic is often times much more flexible, so how [dynamically] do you change the configuration in your network to accommodate for changes allowed for different prefixes?
[FERNANDO]: Well, I don't do that, well it's not it's a customer network but I don't do that because I don't know of real good ways to do it. I will gladly accept any input and usually changing the configuration implies some kind of flapping and I don't like that because I think it's bad for the customers and for the care of the customers, OK? In any case, I have concentrated on getting a stable static configuration. If I need sometimes I need to [ ] it but it's in a long time, not daily based.
Q. [Yes]. So the settings that you have made are bestal true [statistics] or the history of user traffic?
SPEAKER: Well, actually, I can show you this well, this has graphics. I have [Monday] graphics also and that is so, that the traffic is running OK with this configuration. We observe it and we see if we see a long term [TLAENDZ] change it, then we try to fix it but not on the basis of daily traffic?
Q. Yes. OK thank you.
MR. HUGHES: Any further questions? If not, thank you very much, Fernando.
[APPLAUSE]
MR. HUGHES: And I would like to ask our next speaker to come up to the podium. Peter, are you ready? So a familiar face to some of you but have not seen him at a RIPE meeting for a little while and that is Peter Lothberg. Peter is a bit of a [ ] king and has [CLRSs] in his house and things like that and Peter is going to talk to us about running lots of fast networks to various members of his family.
PETER LOTHBERG: A very good morning. And I am going to talk a little bit about what we can do in order to move the IP network closer to the fibre and what kind of next steps we can do and kind of a little bit of a demonstration with this [thank] in the summer.
My mother, who is 75, didn't have the Internet and there was not much to buy on the street that actually worked. I mean, living on a backbone for most of my life I go "what is this?" So I decided to see if we can improve the performance or get her a real circuit. So this is basically an exercise of making many phone calls. Why do this? Well, [any] of the BoF.
Now, here is the real story. I want to give this talk to explain to people here how this is removing yet another layer out of the IP [in]TRA future and most effective and more band which had [VT] and more flexible. If you are working in a large carrier somewhere trying to do this you are going to run into people who tell you it can't be done and it's impossible and they have all sorts of reasons, and most of those reasons coming out that is my job going so they are trying to help you not help them.
In history we have seen a bunch of those. First of all, around near Amsterdam we were told that if you want to build an IP network you shall use X.2025 and then run it directly to [a]DRO mully MREKSers and [they'll] [explodate] Next thing we attached router direct to the WDM system where the aTROP [multiplexes] used to live and they screamed bloody murder. So this is the next step we have to take in order to make the network more flexible and more powerful. It's actually an interesting step because once we get fibre between once we get line path between the two routers, we only need to change what is in the router to make it faster or better or something and it becomes very interesting because not only can you take advantage of advancements, but us changing two things instead of having to rebuild the entire network.
So to get 40 Gig to my mother, what we did was the following: In the old traditional network in a digital network what you do is once you get to the two and a half and 10 Gig, you take router and plug it into the transponder and it's [base] like a border here where there is that is local fibre and then this thing does all the long haul; there is a bunch of optical receivers and transmitters in the path. Now, what you do is you take one receiver and transmitter and put it on your router and you take everything else and throw it away; poof gone. And that way this guy talks to this guy.
Now, there was a lot of people involved helping me here and I just put some of them there was significant, they were [ ] now, first of all you need to install the router at my mother's house and we [even] old VOEK wagon Burkes I greatly increased the [values] I and you have to drive carefully and you basically have to take everything out of the box; it's still 150 kilos. My dad is inspecting, making sure that the son doesn't make too much a mess in the house and you can remember I have made a lot of mess in that house. My mother was instructed to install this thing. I am not kidding. I told her do this and do this and do this because, you know, a phone company doesn't have [ ] people if my mother can't do it, this can't be done right. So we kind of got a mix of different technologies in the garage here and this is actually the back up if the power goes out.
Now, optically, my mother lives here and if you continue outside the picture here you end up in craziness; it's called Norway. So this is about half the way to [OLS] have from Stockholm [ate] 8 fibre rod the road is actually shorter than the fibre rod; it's 100 kilometres shorter. But it's 412 kilometres and our friend at [SOON] just completed their new national research [optimal] super that uses really cool shit that we haven't worked. I went to [SHOON] and said "Can you give me an optical path on your [think] that you are not using." [Anyway], the system they have is so cool. There are connectors where they are on the front where you [wire] up internally. You just need to get the right connectors and off you go and it takes care of itself as long as it kind of understands what signals you are giving it. We are going from the optical from the garage, make sure [ ] can't put the fibre under the router or cut [TSHS] up to here.
Now, we talked about it earlier. We took what used to be here and [transponder]. We just didn't plug it in. We put the router directly. Now this exercise had another challenge in. If you do this in real life you usually have the router standing next to this box so it's a short fibre so you can really ignore the fibre. Now, this is in my mother's garage and this is like 10 kilometres away, 7 kilometres away so the local loop is kind of in the long path. So this piece of fibre, whatever this fibre is, it's going to be added to the performance of the entire system going 400 kilometres. So it actually looked like this, going in here. We have a bunch of amplifiers, and I got this picture back, those two are supposed to attach there and the router is supposed to be here, like I forgot. There is a little bit of dispersion left, don't worry about it, a little bit of dispersion left so we ended up having to put negative fibre. So fibre has this beautiful characteristic that higher wavelengths travel faster and lower travel slower. So that means that you have to get it all back in order and you can make it last. It's either positive or negative, and [normal] glass basically positive and so you can put negative fibre in so we [actually] to add a little bit of fibre and usually you are supposed to put that on the receive side but as I am backwards the system is actually done on the transmit side.
Now, I didn't look at this but the [city] of con STAD they are like open fibre, they will give anybody fibre who, you know, writes them a little cheque and they actually put the fibre in my mother's house. I called them up and said I want to do 40 COMBIG to my mother and I got this lesson [WHAN] this they are suppose and it was triple play. I said "OK, fine, don't worry about it." They measured the fibre for me and I didn't look at this. I was very happy I didn't look at this because they forgot to clean a bunch of [KESHTHS]. If I had looked at this before I had tried, I would have called them up and told them to clean the fibre, probably made no friends but it actually [works], even this, actually this is still there and I didn't bother [to] is it working? I didn't bother to get it fixed but you don't want this kind of you couldn't want to be able to see every connector you have that way.
Now, the [soonest] system looks like this. Here [STOM] home and a bunch of bands and seen then in a system is 8 channels. Basically what they are doing is dropping off 8 channels whereas [university] and then they have to do a little bit more fancy stuff to continue and I cut the [puck] tour because we are going from here to here. If you draw it out and this is what you have to do if you want to figure out if you can do this at home. You have to figure out what is the fibre, what is the fibre attenuation, what is the distance, what is the PMD and the dispersion and then you have to figure out how much amplifiers you have. The coloured thing here, numbers are dispersion [compensation] fibre. This is how much negative fibre we put to compensate the positive fibre and as you see, you start off by making it negative but you come out with this thing so you basically want to shift it off, and the reason you [do] [that] is you want to get you want to get out of the group delay problem. At the end, as this local loop has so much attenuation and the light coming out of the system was actually supposed to plug in a transponder right there, I actually had to put an amplifier here and that sits at my mother's house to get enough power here, so this is not an optimal design and the dispersion compensation to get it right for this receiver is actually sitting there. It was very simple. I got some friends to look at this side and I was here plugging things in and they didn't have any tools there and we only had a cell phone because there was no Internet at my mother's house before we start.
Now here is the amplifier, the [DCU], the dispersion compensation model, the same modules that sits along the line. We had a bunch of them left over from when the system was built so I basically brought a bag of them. They are heavy and there are different distances: 10, 20 and 40 kilometres, and with the right cables you can vary the length of the fibre. Here is the optical plug. They called me up and said "Where do you want to us terminate it?" Do it the way you always do it. We put a triple play box in. OK. Here is the instructions for them to plug it in on the [soon] ET side. I said, "Look, I want you to put the fibre in there" and it actually work. The first phone call was: We can't find your box. It looks it says [Marconi] on it "you are in the wrong [rack]." IP wise, mother's house big router, 10 Gig to [sprint], another 10 Gig to sprint, 10 Gig to the exchange point in Stockholm, 40 Gig back to back, short reach to [soon] ET, so mother could, if she weren't sending, actually fill the circuit up [ ] can't do it.
Now my mother is not, you know, she figured out that the son's clothes could be dried faster by left hanging in there.
Now, power consumption, so I put this [thing] because I had to negotiate this deal with my dad because he could pay for the power because we got all freaked out so I got this power metre, we used 1.56 kilowatt all the time so the cost was €293 per month just to power this thing. So I, after demonstrating that this was working, I go okay but it's really cool. How do I get Internet to my parents without having the big router there? So I scratch my head and go "Oh, there is SFPs like [optical] G, you know, you have all seen them, right, that are actually DWDM colours, so I look at the catalogue and go yeah, there is one thats colour that I am using on this big box so actually I put the G big in there, 200 mail order linking box so I left the fibre from here, pulled the DCM modules out and put it in there and the link came up. So you can do from 40 gig to one gig and change only the end points. I thought that is cool demo. If you take 40 of those and stack them on top and use half of the optical system it's actually cheaper. "
Now, how did this work? This is a typical optical system. Here is the light terminating equipment. Right here is my router. Here is how it's supposed to be. Here is the transponder. Here is the terminating equipment. Here is the OEO transponder, and this is what they sell you. We didn't connect this thing here. We plugged it in here so light comes out here, there is some [muxim] scheme and depending on the vendor, all sorts of different ways, usually these are called interlevers or something, but bottom line is you get the window like a radio channel out of light cut out for you and it's either 50 gigahertz or 100 gigahertz wide and long hall systems are usually 50 gigahertz wide. Now, if you compact your channels narrower, of course you can stick a bunch of channels into that 50 gigahertz. Somehow they take those 50 gigahertz channels, 80, 90 [slicks] or something, and they pack them all together and they put them out on an amplifier and they put all sorts of stuff here that they want to sell you, stupid things, but basically you put amplifiers to compensate for the loss in the fibre and then you filter the parts. There are actually two in that system, one going this way and one going that way.
What happens to light in fibre is kind of funny. First of all, optical network is the slowest network you can every buy because it's only 0.64 of speed of light in vacuum, so sending singles through the heads of the people is much faster.
So the fact that the pulses, the higher wavelengths travel faster than the lower wavelengths, what is going to happen is the signal that you put into it that is not perfectly shaped called a [soliton], it has different colours even if it's short narrow band thing, so there is some different frequencies going out, the signal that comes out of your router or your thing has different colours and they travel at different speeds. So what is going to happen is that lower frequencies are going to come out after the higher frequencies. So as you travel along the fibre path, the pulse gets wider and wider and wider. Now is this bad or good? It's actually a problem because once the fibre gets too long, you get continuous light so you have to do something about it. There is two ways of dealing with the problem, actually three ways of dealing with the problem.
You can actually put fibre in the path that is negative instead of positive, so basically you put fibre in that has a reflection where the shorter wavelengths run slower and you can actually manufacture so if you are 100 kilometres out in the field you probably only need 9 kilometres of fibre from the spool. So that is why optical network, when you compensate for PMD, gets longer.
So it's also a little bit of a feature because you really don't want the perfect analogue system. You want to push it out in the non-linear region a little bit in order to not have all those channels to intermodulate with each other. It behaves like, you know, you tune your radio and you can hear two stations at once, and that is not really what you want.
Optical signal to noise: Basically, this is like copying tapes, like this is radio or this is analogue stuff so every time you amplify, you add noise. And basically the game is to try to keep as average power across the line as possible. That is why submarine cables has a bunch of many, many more amplifiers so they can keep the power levels as equal as possible.
So if you have like 100 kilometres of fibre, you have 20 DB loss, you put a 20 DB amplifier in, you have, you know, 600 picosecond or of dispersion you put the dispersion compensation spool in the middle of the amplifier because you can't put it on the input because you lower the input signal, you get more noise. You can't put it on output because there is too much power and the fibre goes numb.
Can we go as far as possible? How far can we go? Well we are limited by the signal to noise ratio. If you are looking an optical spectrum analyser between noise tour and the actual signal, there has to be some signal there. Now the good news is that as evolution started to take off again after the market came out of nuclear cold war, people are actually starting to develop more and more cool shit sorry cool equipment. So, what happens is that the signal to noise ratio you will actually need for even bandwidth drops. This is like modems, they are getting better and better every day.
Now, there is a bunch of things that happens, get modulated because of all the nonlinear effects and what you can do is you can put filters in or try to shift the dispersion. Now, LSNR, if you want to calculate, is something like this. It is the power density, the input power of the amplifier, the noise factor of the amplifier, the number of amplifiers and the number of optical channels. So here you see the number of channels comes into play. You can have one channel and you can go longer than with many channels. So if you have many channels, you need more power.
Now, there is another thing that wasn't in the old system, it's called forward error correct. This is like a [RAID] disc, right. You add extra data to the bits and that way you can correct for errors. And it's kind of cool. They standardised the payload and they standardised how the container looks because you want it to have a way to monitor the signal. If you just have an ethernet as it is on a computer or router, there is no management; you can't see what is going on. So what you do is you take any kind of signal and put it in a wrapper and the wrapper then has the same kind of overhead a [sonnet] has so you can get better rates and power management and all that stuff.
So there are containers called OTU and there are three of them standardised and they are arguing if it's going to be 160 gig or 100 gig and there are other people in this room that know more about it than I do. But basically this is OTU free and the stuff we were playing with in the mother's garage uses this kind of encoding because it's standardised and doesn't do anything fancy. It's just that in the FEC fields are not doing what G.709 specifies. We actually try to do something better.
Here is what an uncorrected transceiver will do and here is the gain that you get with OTU 709 encoding or enhanced encoding and you see there is about eight and a half DB of gain, but also look in an unencoded thing, the curve goes like this. So if you are starting to get OSNR or your fibre goes bad or you are approaching the maximum distance, it goes slowly down creepy on you. If you are running on FEC, what happens is you get to the point where you can't correct and it and goes 'poof!'. So if you are trying to figure out if the circuit is going bad, it's basically going to say 'poof!' on you. The cool thing is you that you can watch how many errors you have corrected per second to figure out where you are on the slope going down or how close you are to the cliff.
There are a lot of problem with fibres called PMD. Fibre is supposed to be round. Old fibre is squared. So the distance when when light goes this way, it travels a shorter distance than when it travels that way. So when the light crumbles along the wire or actually turns around because of the mechanical imperfects, what is going to happen is that you will get the signal that went that versus the signal that went that way are actually going to come out at different times the same way we have dispersion but now it's caused by the fact that the fibre is not round. This is a real killer for old fibre and it's measured in picoseconds per square banan or something, like a picosecond per kilometre half and basically if this number gets too high, you have a problem.
There are ways to deal with it. You can have a radio receiver, a quadrant receiver on the receive side and can recover the polarisation as part of your demo relation problems but there are limits to what you can do, or you can do optical PMD compensators and you can do the multistage and do them along the line but now it becomes complicated and in a lot of cases it's probably better to just replace the fibre.
Another thing that happens on a system like this is that you have a bunch of carriers like on your FM radio, what is going to happen is that the harmonics of them will all intermix and as we are traveling down this fibre that is not perfectly linear, they are going to be added and modulated to each other so, as more channels you put in, you are actually creating small little, you know, channel carriers that all will add up so you will get a complete spaghetti. So the way you can deal with that is basically you can space your channels unequal so the beats end up with the filter dips or you can try and use chromatic dispersion, basically shift the fibre off into the other side of the nonlinear and that way minimise those. But this is the more channels you get, the more headache you get.
XPM is basically just fibre. Depending on how much light there is already, it changes its translucency how the attenuation changes a little bit. So you actually get something that AM modulates one signal on top of another signal and the way you can deal with it is basically again try to shift the dispersing compensation along the lines so you are off perfect and lower your spacing between carriers. Submarine system uses usually longer spacing than 50 gigahertz used on land systems.
Now, up 'til 2000 we were running Morse Code da di da, da da di da on this fibre, we were turning the light on and off and suddenly people go we want more bandwidth and development took off again. These are acronyms for different kinds of modulation schemes and NRZ is what we have in all the other systems 10 gig systems and people started do all sorts of fancy things. You look at those, they get to about 2 bits per hertz of optical band which had: I want to TUSH ET so I need three times better than all of these people so none of this is good enough.
There are 3 ways you can moderate, you can change the amplitude, you can change to phase and polarisation. Those are the things that you can do to fibre and then the fibre helps you and does some of it by self so you have to be able to separate what the fibre did versus what you put in at the end. The modulation scheme used to in my mother's house is do binary basically instead of turning light on and off we have three levels of power and we basically by having 3 levels of power we can take this the 40 gig bit street and comes 20 gigabits bandwidth SOTS the blue signal and it will fit inside the black thing WHAS the filters. Who is keeping track of time. Nobody had said slow down yet. Some of those encoding schemes have very particular effects. Like, CS R Z has those two Mickey Mouse ears with smiling on the sides. You have to make something that fits inside your filter. So my blue signal here which is what we did to my mother fit inside the filter. The green thing is the DPS casing, where it actually has better properties because the power levels are slower. It's actually enough for what fits in here to make it work. The red one is what you would get if you were to take the old ten gig enKOEGD, it just doesn't work right. Here is a cool thing that we discovered: The optical transport system, we have something that sits there and monitors the powers of the channels and balancing them to make it all perfect or add goose as possible, there is a mod that looks at the signal and makes a guess on what it's looking at. It assumes that they are this kind of signals. If you feed it something like this and manage today NIEND peak that, the power level it would set even if it was supposed to set is that power level so. You have to be a little bit careful when we start to invent a new modulation scheme that doesn't like would expects because they get no markers.
In DPSK you basically change the face instead of just the amplitude so you get 3 DB more at the same power level and your receiver becomes more complicated, basically you have to have a delay loop with a piece of fibre, the transmitters like as the same as a DPS do you know BIEN REESHLG, free DB. STIEMGS not. This is the other one that is out there and it's available and there are boxes you can buy. Now, there is a bunch of other schemes that were on my slide and I made this stable, here is the same optical bandwidth. How much power do I need to receive the signal? What is the PMD tolerance for the different encoding scheme, and how much KROM attic dispersion does it handle and then kind of how much does it take to build this. Now, so if you want to build something we want to use the we have to make sure that the PMD we have on the path is lower than that, the PMD or the dispersion is within this window and if it's not we have to fix it somehow. This is this is a very hard number in old fibres to deal with. Now, if you go shorter distance you can always you can have higher this and lower that and so on so there is not a fixed you can't really say if it's going to work or not because PMD and do all sorts of tricks. So now, I put THECHL in, you can pull the presentation down from the web page but I put in the other modulation methods that are on the chart, how you actually go build them and the latest and greatest that we haven't SNEEN real product is this polarisation PMD QPSK, they are having four different phases, four things they are playing with so every bit stream symbol rate is only 10 gigabits to get 40 gigabits of transport. That means your optical system and characteristics that is 10 gig signal even if it carries 40 gig data. This is the 1,200 go to DASH and it continues but adding phases. To demo like this, you have to basically build a radio, so you have to have a local OSH later, you have tomorrow IQ deMODZ and it becomes complex and we are basically stretching out the technology what we can do on the demo later side. None of this is good enough or TESH ET a funny chart that shows where you can put fought on I can stuff and electronic stuff so there is a lot if we can get 8 and a half DB of OS by forward error correct the signal. 3 D INTCHLT doubling the distance, it's a lot. So it sounds like a small number, it's a great number. There are more than technologies for soft decision FEC code, where you can get about 1 DB. I don't know if the road ends there, if somebody can come up with something better. The fact that you know you are moving packets should be a way to feed things back into your forward error correcting soft decision to say look, it's supposed to say 4 here unless they are cog IPv6, first time I mention IPv6, my mother has IPv6. Now here is another cool thing. If you are looking at a traditional transport system where you are the client, and here is where it stops working, you don't really know anything so when the long haul thing starts to get to the point where the transponder in your 10 gig system no longer can do it, it basically sends LONS of frame, light goes out or you get loss of frame and you didn't know that is going to happen, it just happened on you because something happened to the firer plant right. If it's all built into the router, what you can do is you can sit and watch the FEC limit, basically you watch how much errors am I correcting per second and when you know you are to go here, we basically tell the router we tell the routing protocols to switch so before you know that it's going to go back he will switch the circuit out. If you are having heavy PMD on a network it looks like it took bunch of and you are looking through them with bin on collars, at some point going to be in the wrong position and the circuit stopped working but only Hans maybe once a year so you can actually run at much smaller margin by doing something like that by doing that, if you have this kind of network you have to make enough margins so it never happens.
On a 40 gig transponder into the fibre on to the long haul system, this is the power that goes on the fibre that leaves the building, we are launching at the same power, the window on the receivers are about the same size but the 40 gig trance receiver needs a little bit more power: The OSNR requirement for doing do you know BIENry is about 18.6 DB versus 15 DB on the 10 gig thing that you already have so you see this is three times worse but we already had the FEQ that was better than this so USHLG comes out, if you can do 10 gig you can do 40.
The PMD tolerances, if you are doing 10 gig STAMS plus/fine MUS 2000 pack seconds. And 10 on a do BIENer receiver you have to be within plus/minus 150 pack owe seconds. It turned out this was too hard to make people understand because the systems were built, if you want to do this you only need one set of dispersion compensation fibre for your entire 80 channels. If you want this you want to have you have to compensate each individual channel. The to my mother they have nothing in THOEM deal with this but what people figured out was OK this doesn't work so actually on things that you can buy, the PMD compensators are built into the cards so you can have plus, minus 700 here so USHL you can deal with it without having to have external fibre. In the case of my mother we have to have experiment dispersion compensation sitting in the line on the receiver side just for that channel so fix to clean up what was left over from this.
If you are looking at the dispersion map along a wire, you basically are building up dispersion in the fibre that is in the ground which is called SMF 28 in American terms and G something 652 in Europe, and then you have dispersion compensation fibre and long haul so you are trying to keep this at zero and at zero when you exit your system. Now, if you are playing around, I say the receiver has 150 packet SOEKDZ plus minus window, if you are looking at how good the receiver works you can see that it's not actually working best at, when it's perfect. If you have off a little bit that is your best receiver is and this is about half a DB of gain so you can actually – there is a little bit of margin here, if you want to make a system that just works by itself you have to look at it from this level but you can if you want to optimise make sure it sits on that peak.
So here is your optical system, they already have compensation sitting here if, that is not enough, there is a XEN sitter that sits there. If that is not enough you have to add a SPOOFL fibre, the way you calculate the dispersion through in here, and that is about it. Now, the noise figure then we already saw. Here is the OMENTical spectrum of a system that has 6, 10 gig signals and two 40 gig signals. Here they are launching the 40 at LOUR power just because they were getting interference. This is pretty and here you see that what I mean that OSNR is the distance from here to here, basically how much signal do I have right from, here to here at 01 Nanometres 9 width.
So we are going to build a path. We have to have the fibre character STSHGS the first thing you are going to run into they don't know where they have, you have to ask your transfer to keep giving it to me or measure if you have to get the fibre TAVENLT the first thing you have to get is fibre data and once you get that you can figure out all those numbers we talked about and figure out am I within the ranges that the piece of equipment that I have company handle? You have to know the loss and distance, right? You have to figure out how much dispersion compensation they already put in, what kind of fibre it is. You have to figure out how much PMD there is, how you compensate for that and if you want to get better OSNR you can have narrow band filters but that limits your channel band WIT it's not what you want to do if you want to go 40 gig. You can put ram a.m. MRI fires, off fibre that is in the ground on the receiver you put the big lacer and shine lines towards the transmitter RS you are using the fibre in the grounds as an amplifier and placing our hands like now the power level along the fibre becomes more equal because it basically goes down and starts to see that the pump light you put into the fibre and the gain goes up again.
Here is a little graph that shows on the do BIENry card that we use today my mother, here is kill MEEFT,. In theory could you do something like 1,700 kilometres with this. Now, that is left F you have a peeves fibre and you have no slope compensation, you have no TDCs and PD no TDC and no PMD on the firer this is the distance you can have it. 1,100 kilometres. So the green thing here is the dips tans you can carry at this wavelength. Once you change wavelengths you kind of lose.
If you do no slope compensation but you have a TDC and your card has plus or minus 700 pack 0 seconds that is the number of charges you can do at the disRANS distances so life became better. If they do slope compensation i.e. they are trying to make this thing as flat as possible you actually get a little bit more of distance so this is what you can do on paper, no margin, using do BIENry encoding and very few people have fibre stretches that are in the 1,500 kilometre range. So it's worthwhile trying.
This shows the pen AFLT going through filters. Basically every time if you want to go very long distances you will find out that oops, there are two systems so you want to go from one system to another, once you go from one system to another you go from one set of filters on the first to one set on the second system so basically you are filtering your 50 gigahertz channels so it becomes narrower and that ends up as OSNR , inter lever is basically the WDM people name for the filter that sits there that separates out the channel.
I put this graph in that shows the distance SA where the FEK margins is with different kind of encoding systems that we talked about and we are doing that PBL this is the one we use today with my mother and you see that once we get to about 800 kilometers we are starting to use the error correct but it may work up here. Your experience may vary but use this as a guideline looking at your own the data you GRET your fibre people or transport people look at this graph and see if you are reasonably within here. It may be fine.
Now the bash key people told me 400 kilometres, so the longest run SHOON ET has in speeder is lieu lay STOM home. It's this A fibre, metres of fibre and we had to add extra spool of fibre to get the dispersion compensation right so the total distance is 11 1136 kilometre which is way up on that scale. STOM home MUN I can't CHT there, 50 amplifiers, where you drop off bands, there is an extra amplifier in Stockholm because that 46 kilometres of extra fibre costs attenuation and the power level wasn't big enough to get into the system and make it work happy so I stuck another amp in there so it's not really what it could be.
Now lose lay is far away. So it was a four day product dropping the router up by the VOEK wagon bus and this thing had TLIPD two times, I think it was 2800 kilometres. For people living here in the Netherlands. Sweden looks like this, there is nothing.
OK. Stockholm. Starts here. Channel 5, all those amplifier and all those fibres, all across the system over here, the whole damn thing, so we go as far as I could. Plug and play. Here is the dispersion map. And you see that going north to south they actually, the people who built the system manage today come out at 0, so north to south is fine, you don't need to do anything. South to north, i.e. going up, they left the 1000 pack 0 seconds, I don't know why. Couldn't figure that out. But probably because they are modules were increments of something. So we have to get rid of this piece and that was done by putting fibre on the transmit side. Why did I put fibre on the transmit side? Well, this receiver is a LOOUN is thousand km away, I did not have 50 kilometres worth of extra fibre because I wasn't thinking again. So when I found the fibre I was back at Stockholm and you don't want to fly back and forth, you put it on transmit to see if it works and if it works you leave their T there so. Add that. Now, they call this Peter installs, it's a mess. Here is the boxes with extra fibre to do dispersion compensation plus and here is the dispersion compensation minus extra modules, here is the amplifier, here is the router and it plugs in there somewhere and there is a 40 gig short reach card that goes to SOON and a bunch of 10 gig stuff that goes to other people as you saw on the network drawing. And they had put a tag there, this is Peter's fibre.
This is the system in LOOUN and you plug in here and make sure you don't make any mistakes because then they scheme ought. Put the router here. My mother has counsel graded to big bit, same set up and it just worked, almost.
Now, that is it. OK. Thank you very much.
MR. HUGHES: Any questions? That was bloody fascinating if I might say so. More about optical networks than I think anybody can stand before coffee break. We will make sure it's in the afternoon in future so we don't ably our minds. Do we let people go early? So you get to go early. Go into the coffee break and back at 11:00. So you get a nice time to go and socialise in the corridor thank you.
(Coffee break)
|
|
|
 |
| RELATED TOPICS |
| RESOURCE LINKS |
|
|
|
|
| |
|
you
are here: RIPE -> RIPE
Meetings -> RIPE 55 
