第三讲ishier
- 格式:ppt
- 大小:2.09 MB
- 文档页数:3
首先,IS-IS是OSI(开放系统互联)协议簇的一部分,OSI使用称为CLNS(无连接网络服务)来提供数据的传输,因此,IS-IS也就是以CLNS为基础,包括IS-IS协议的路由运算也是以CLNS地址的方式来运算的,,CLNS对地址的表示方法与IP地址的表示方法存在着根本区别最初的IS-IS采用CLNS地址来工作,然而整个互联网的标识与运行都是使用IP地址,那么要使IS-IS真正用到互联网中,就必须使用IP地址为基础,正因为如此,在原始IS-IS的基础上,加入了IP地址,使得IS-IS能够认识IP地址,并且对外显示出IP地址的信息,这种具有IP地址信息和功能的IS-IS,称为集成IS-IS。
(注意:我们现在所讨论的所有都是集成IS-IS)可以把集成IS-IS比作是电脑,电脑都必须有主板、CPU、内存、硬盘,显卡才能正常工作,但是要把这些东西全部都组合在一起,可以组成台式机或者笔记本电脑,台式机的机箱可以各不相同,笔记本的外观也各式各样,我们拿到一台电脑,不管看上去是什么样的,但它内部都有主板、CPU、内存、硬盘,显卡来作为工作的基础,而IS-IS协议的工作就像电脑的工作,不管让IS-IS表现出什么效果,让人们看到IS-IS长什么样,但在IS-IS内部核心,始终是以CLNS为基础来工作的,就像电脑的主板、CPU、内存、硬盘,显卡,这是少不了的,我们用户也不用看到,而IS-IS支持IP地址就像电脑要用什么机箱或者要装成台式机还是笔记本一样,只是让我们看到一个形式,让我们容易分辨而已!我们对集成IS-IS 的操作仍然像操作RIP、OSPF一样对它进行操作和配置!IS-IS支持以太网(LAN)和帧中继网络(NBMA)ISIS和OSPF一样,属于链路状态协议,可以在网络发生变化时快速发现和收敛同样支持VLSM,比较灵活!IS-IS对路由的metric计算是用接口度量来计算的,一个接口的度量最大值为63,如果一条路由经过了多台路由器,那么就把接口度量都加起来,但一条路由的总路径度量,也就是最大度量不可大于1023。
L2-U3-1/4 listeningBus schedule 1 / about EarthA1.Here is a bus schedule at a bus stop.2.It has schedule for three buses between eight and nine thirty in the morning.3.Bus No. 38 has a regular schedule.4.It comes every fifteen minutes .5.Eight ten, Eight twenty-five, eight forty, eight fifty-five, nine ten, nighttwenty-five.6.Bus No. forty-seven comes less often.7.It comes at eight twenty, eight forty-five, nine twenty-five.8.Bus No. sixty is the earliest bus.9.It comes at eight o-five, eight thirty, and nine o’c lock.10.Bus No. thirty-eight has stops at the main train station and the airport.11.Bus No. sixty stops at the main train station, but does go to the airport.12.Bus No. forty-seven doesn’t go to either the main train station or theairport.B1.This is our planet earth.2.It has shape of a sphere, like a ball.3.The equator(地球的赤道) device the earth into two halves equator.4.The northern half is the northern hemisphere(半球体).5.The southern half is the southern hemisphere.6.Europe and Asia are both in the northern hemisphere.7.Australia and a third of Africa are in the southern hemisphere.C1.The earth moves in two ways.2.First, it’s been and rotates.3.It’s rotates around its axis.4.The axis is a line through its center.5.It takes the earth twenty-four hours to complete a rotation.6.That is the length of one day.D1.Second, the earth travels around the sun.2.It takes about three hundred sixty-five days for the earth to go around thesun.3.That is the length of one year.4.It’s average speed during the year is about thirty kilometers per second.E1.The earth’s axis is not perpendicular to the sun.2.The earth’s axis is tilted.3.It is tilted at an angle of around twenty-three point five degrees.4.This tilt causes the seasons of the year.F1.During part of the year, northern hemisphere is tilted toward the sun.2.When the northern hemisphere is tilted toward the sun, its summer.3.When the northern hemisphere is tilted away from the sun, its winter.4.This is why the seasons are opposite for the northern and southernhemispheres.5.When it’s summer in Australia, it’s winter in Europe.G1.What would happen if the ear th’s tilt were different than it is now.2.If the earth tilts were larger, summer would be hotter.3.If the earth tilts were smaller, summers would be cooler.4.If the earth had no tilt at all, there would be no seasons.5.Therefore, the angle of the tilt is very important for life on earth.H1.The earth is ninety-three million miles from the sun.2.The sun is at the center of our solar system.3.Our solar system has several planets which orbit around it.4.Our planet is the third planet from the sun.I1.If the earth were closer to the sun, our planet would be hotter.2.If the earth were further from the sun, our planet would be colder.3.Therefore, the distance between the earth and the sun, it’s very important.练:1.We have to wait until 9:00, which is ten minutes from now.2.Whi ch bus doesn’t go to the main train station.3.The northern half is the northern hemisphere.4.The length of one year is about three hundred sixty-fix days.5.What causes the seasons of the year?6.In winter, the earth tilts away from the sun.L2-U3-1/4 vocabularySeasons / difference disease1A1.Winter is the coldest time of the year.2.Winter night are long and the days are short.3.Summer is the hottest time of the year.4.Summer nights are short and the days are long.5.Spring comes after winter and before summer.6.Spring is when trees turn green.7.Autumn or fall comes after summer and before winter.8.Autumn is when trees turn many colors and leaves fall to the ground.9.Some countries have a rainy season.10.In rainy season it rains almost every day.B1.This boy has a broken leg.2.He can’t walk without crutches.3.This girl has a sore throat.4.She doesn’t feel good because her throat hurts.5.This girl is vomiting(呕吐)6.She ate something bad, so she’s throwing up.7.This boy has a fever.8.His temperature is thirty-nine point five degrees Celsius.9.This girl has a headache.10.She doesn’t feel good because the headaches.练:1. In spring the days become warmer each day.2.He broke his leg in a skiing accident3.He needs to drink plenty of liquids.L2-U3-1/4 dialogueTalk about seasons / take bus to train stationA1.I really hate this cold winter weather.2.Really, I don’t mind it, it’s not so bad.3.First, it’s really cold.4.Second, I don’t like the short days and long nights.5.It’s always dark.6.Sure, but what’s the else thing?7.Spring is coming and warmer weather.8.Yes, and longer day.B1.What about summers?2.Do you like hot weather?3.I don’t like summer weather too.4.Is it too hot for you?5.Yes, it’s too hot.6.Every day is hot and humid(潮湿的).7.So, I always feel tired and thirsty.8.Yeah, me too.9.I take a shower twice a day.C1.So, which is worse, summer or winter?2.Good question.3.I guess I prefer winter.4.What about you?5.Well, I prefer winter too.6.I don’t mind the long night.D1.Of course, the best seasons are spring and autumn.2.Which those do you prefer?3.That’s between those to, I prefer autumn.4.Why do you prefer autumn?5.I love it when the leaves turn different colors.6.I agree with you.7.I guess autumn is my favorite season too.E1.I surprise, Tom.2.I thought you prefer spring.3.Why is that?4.Everything is young and new in spring, right?5.Well, maybe I’ll change my mind.F1.Excuse me, when is the next bus?2.To where?3.To the train station.4.Let’s look at the bus sc hedule.5.Which bus goes to the train station?6.Buses forty-eight and sixty both go to the train station?G1.It’s eight fifty-nine now. So, I just miss the number thirty-eight.2.Right, it come a few minutes ago.3.The next bus to the train station comes at nine o’clock.4.Yes, I hope it isn’t late.5.The traffic is heavy.H1.Oh, look, the NO. sixty is coming now, right on schedule.2.Great, thanks for your help.3.You’re welcome.4.Have a good day.5.You too.练:1.The weather gets warmer in spring.2.Winter has the shortest days and the longest nights.3.She takes a shower twice a day.L2-U3-2/4 listeningDen’s workday / Ben’s lunchA1.Den is a pilot.2.He flies airplane to all parts of the world.3.This was his schedule yesterday which was Sunday.4.He woke up at six-thirty.5.After waking up, he got out of bed and brush his teeth.6.After that, he took a shower.B1.He and wife ate breakfast together at seven fifteen.2.He had a big breakfast of eggs, potatoes and fruit.3.They finished eating at seven thirty.4.After eating, he put on his uniform.5.At eight o’clock, they left their house, and his wife drove him to theairport.6.The traffic was heavy, so it took forty-five minutes to get to the airport.C1.They arrived at the airport at eight forty-fifteen.2.After going through security, he checked the weather along the flight path.3.At nine-thirty, he boarded the plane.4.He and co-polite talked about the flight.D1.By ten-thirty the passengers are all on board.2.They pushed back from the gate at ten forty-five.3.From the gate it took ten minutes to go to the runway.4.There were two airplanes in front of them, so they had to wait.E1.They finally entered the runway at eleven-ten, which was five minutes behindschedule.2.Two minutes later, at twelve, they took off for Beijing.F1.The flight from San Francisco to Beijing took twelve hours.2.They landed in the afternoon one day later.3.In Beijing, it is already Monday.G1.After leaving the aircraft, he took a bus to hotel in Beijing.2.He checked in at six pm and went up to his room.3.When he got to his room, he took a shower.4.Then he went downstairs ate dinner.H1.After eating, he went back to his room and watched some TV.2.At ten o’clock, he went to bed and fell asleep.3.The day after tomorrow, he’ll fly back to San Francisco.4.His wife and children will be happy to see him.5.They always miss him when he’s away.I1.Ben often buys lunch at school.2.He usually spends around five dollars for lunch.3.Here are the prices of some of the things on the menu.4.Sandwiches are one dollar fifty cents.5.Ben likes sandwiches but yesterday he had two slices of pizza.6.A slice of pizza costs the same as a sandwich.J1.He also had some fruit,2.He had banana, which was a dollar.3.He was hungry so he also had a salad.4.The salad costs one dollar twenty-five cents.5.He had a bottle of water from home, so he didn’t buy anything to drink.6.Altogether his lunch cost five dollars and twenty-five cents.7.He enjoyed his lunch, because the pizza was good.练:1.What was the total cost of his lunch?2.After going back to his room, he watched some TVL2-U2-2/4 vocabularySeveral vehicle(交通工具) / difference disease2A1.Airplanes are the fastest way to travel a long distance.2.Airplane travel is one of the safest ways to travel.3.This is a truck or lorry.4.Trucks carry and deliver many of the things we buy and use each day.5.Buses are a good way to travel in a city.6.A bus can carry many passengers and is less expensive than a taxi.7.Buses can carry more passengers than a taxi or trunk.8.Ships transport freight, such as oil or automobiles across the oceans.9.Some ships carry passengers on ocean cruises.10.Ambulances carry sick or injured people to a hospital.11.If someone is in an accident, call for an ambulance.练:1.One very famous passenger ship was the Titanic.B1.This boy has a cut on his finger.2.He cut his finger with a knife, so now it’s bleeding.3.This girl is coughing(咳嗽).4.She is coughing because she has a bad cold.5.This boy has diarrhea(拉肚子)6.He ate something bad so his stomach hurts.7.This person is dead.8.He is dead, because he was hit by the car.9.This man is drunk.10.He’s drunk because he drank too much wine.练1.He has to go to the toilet again and again.2.He wasn’t careful crossing the street.L2-U3-2/4 dialogueLisa’s question about earth / her lunchA1.Can you help me?2.Here’s my homework question.3.Sure, I’ll help you if I can.4.What’s the question?5.What would happen to the seasons, if the earth stopped to rotating.6.Stop to rotating?7.You mean the earth would always face the same direction?8.Right, so a day would be a yearlong.9.Wow, that’s a difficult question.10.So, each season would be one fourth of the year.11.I don’t know, I am not sure.B1.During the summer, the sun would always be up.2.It will be very hot.3.And during the winter, it will always be dark, right?4.Yes, I think that’s right.5.Maybe there wouldn’t be any spring of fall.6.Hey, I don’t know.7.Ok, let’s check that on the internet.C1.Hey, why are you sitting there?2.Where’s your lunch?3.I left it at home.4.What’s going to the cafeteria and get lunch together?5.No, I can’t.6.Why not.7.I don’t have any money.8.I left my money at home.9.That’s ok, I have enough money.10.You can pay me back tomorrow.11.Thanks, I’m hungry.D1.What would you like?2.That pizza looks good.3.I had pizza yes terday and it wasn’t very good.4.It didn’t have much taste.5.How about the sandwiches?6.Sometimes they’re ok.7.But hot dogs are usually good, and never bad.8.So, what are you going to get?9.I think I’ll have a hot dog and a banana.10.I’m tired of pizza and noodles.E1.What about something to drink?2.Oh, just water.3.The soft drinks have too much sugar.4.I don’t want to gain weight.5.Ok, I’ll have the same thing.6.Great, so the total coast is how much?7.Two hot dogs, two bananas and two bottles of water.8.That’s six dollars.F1.Do you have enough?2.No problem. I’ve gotten ten dollars.3.Thanks again. I’ll pay you back tomorrow.4.No problem, don’t worry about it.G1.My mother isn’t going to be happy.2.Why is that?3.She made my lunch and I left it at home.4.Oh, I see. My parents like that too.L2-U3-3/4 listeningLisa got a cold / a man got an accidentAst night, Lisa come home later than usual.2.It was raining, and she didn’t have her umbrella.3.When she got home, she was cold and wet.4.She was tired and she didn’t feel well.5.She got out of her wet clothes and went to bed.6.She just wanted to sleep.B1.This morning she woke up with a bad cold.2.She had a headache and a sore throat.3.Her mother took her temperature.4.Her temperature was thirty-nine degrees, so she had a fever.5.Her mother told her to stay in bed.C1.Lisa sore throat hurt a lot. So, her mother called the doctor.2.She made an appointment for eleven o’clock.3.Lisa got out of the bed at ten o’clock and got dressed.4.At ten fifteen, they left their apartment.5.It was still raining, so they took a taxi.D1.They got to the doctor’s office at then forty-five.2.Her mother gave Lisa’s name to the receptionist.3.Then they sat in the waiting room.4.They waited for ten minutes.5.Then a nurse came and took Lisa into another room.6.The nurse weighted Lisa and took her temperature.7.Then Lisa waited for the doctor.E1.When the doctor came, he looked down her throat.2.He listened to her heart beat.3.Then he gave her a shot in her arm.4.The shot hurt a little, but Lisa didn’t mind.5.She wanted it to help her feel better.F1.After leaving the doctor’s office, they went to a pharmacy.2.Her mother bought some medicine.3.Then they took a taxi home.G1.For the rest of the day, Lisa stayed in bed.2.She took some medicine and drank a lot of liquids.3.She slept for a couple of hours.4.Then she listened to some music.5.By six o’clock, she was feeling much better.H1.Now, she’s thinking about tomorrow.2.She’s looking forward to going to school.3.She doesn’t want to stay home again.4.But she may have to stay home.5.Sh e can’t return to school until her cold is gone.6.She doesn’t want her friends to catch a cold.I1.Yesterday, there was a terrible accident.2.It happened in front of a subway station.3.There was a crosswalk, and the light was red.J1.Cars were coming from all directions.2.A young man wanted to cross the street.3.He didn’t want to wait for the light to change.4.He looked both ways and then started to run across the street.5.But he didn’t see one car and it hit him.6.He flew up into the air and came down on the road.7.His head was injured and he was bleeding.K1.Several people used their phones to call for an ambulance.2.It arrived a few minutes later and took the man away.3.We still don’t know if he lived or died.4.Hopefully, he’s alive and will get better soon.5.So be careful when you cross the street.L2-U3-3/4 vocabularyWhat people need/ water’s difference stateA1.Everyone needs food and water.2.Without food and water, we not live.3.People need a place to live in and sleep.4.We need a place to keep us dry in rainy weather.5.We need good heath to keep us strong.6.Daily exercise is a good way to stay in good health.7.We need skills to find a good job.8.Without good job skills, we can’t keep a good job.9.We need money to buy things, such as food.10.Without money, it’s very difficult to have a good life.B1.Ice is the solid state of water.2.The freezing point of water is zero degree Celsius.3.This is the liquid state of water.4.Liquids take the shape of their container such as this glass.5.Water vapor is the gas state of water.6.Water becomes a gas at a hundred degree Celsius which is its boiling point.7.We use the scale like this to weight things.8.One kilogram is equal to two point two pounds.9.This is a thermometer.10.We use thermometer like this to measure temperature.L2-U3-3/4 dialogueTalk about Angela / talk about traffic accidentA1.I didn’t see Angela today.2.Did she come to the office?3.No, she didn’t.4.She stays at home, she sick.5.Sick, what’s wrong with her?6.She’s a bad sore throat and a headache.7.She’s staying in bed.8.Oh, that’s too bad.9.She looked fine yesterday.10.She was fine yesterday, but last night she got wet in the rain.B1.Didn’t she have an umbrella?2.No, she didn’t, so she really wet.3.It’s a long walk from the subway to her home.4.She didn’t get home until late.C1.It was really raining last night.2.I got wet too, and I had my umbrella.3.It was that strong winds.4.My umbrella was not much help.5.Still, it was better than nothing.D1.Anyway, when you see her? Tell her to get better soon.2.We miss her at the office.3.There is also an important meeting the day after tomorrow.4.I hope she can be there.5.Ok, I’ll call her and let her know.6.If she can’t be here, we could set up a conference call.7.So please let me know.8.I will.E1.I saw a terrible accident yesterday.2.What happened?3.I was waiting to cross third street and the light was red.4.Yes, many people don’t want to wait for that light.5.It takes a long time for that light to change.6.So, there is a young man did not want to wait.7.He ran out into the traffic, and a car hit him.F1.So, he didn’t see the c ar?2.Right, he didn’t see it, and it was going too fast to stop.3.So, what happened to him?4.The driver got out of the car and stayed next to him.5.He was just lying there.6.I went out to direct traffic and several other people called for an ambulance.G1.How long did it take for the ambulance to arrive?2.It didn’t take along.3.I think it only took ten minutes.4.How was the young man when the ambulance arrived?5.His eyes were open, but there was blood coming out of his mouth.6.It didn’t look good.H1.Did the police come?2.Yes, they got there just before the ambulance.3.Did they ask you any questions?4.Yes, they did, they ask a lot of questions.5.So, I told them what happened.I1.What about the driver of the car?2.When the ambulance left, he was sitting in his car.3.The police were talking to him.4.What happened after that?5.I don’t know, because I had to get home.J1.How did you feel?2.I kept thinking about it.3.From now on, I’ll be more careful crossing the street.4.Me too.L2-U3-4/4 listeningComparison people’s weight and height/Cathy’s one dayA1.Here are four people.2.Two of them are tall.3.The man on the top left is very tall.4.He is six feet three inches tall, which is about one point nine meters.5.The woman on the top right is five feet eleven inches tall.6.She is tall, but not as tall as the man on the left.B1.The girl on the bottom left is short.2.She is less than five feet tall.3.She is four feet ten inches, which is about one hundred forty-sevencentimeters.4.The boy on the bottom right is taller than the girl.5.He is five feet one inches tall. Which is about one hundred fifty-fivecentimeters.6.He is three inches taller than she is.C1.Now let’s look at the weight.2.The man on the top left is the heaviest.3.He weights two hundred twenty pounds which is a hundred kilograms.4.The woman weights less than he does.5.She weight a hundred and thirty pounds which is fifty nine kilograms.D1.The girl on the left is very thin.2.She doesn’t weight very much.3.Her weight is just seventy pounds.4.The boy is much heavier.5.He weights ninety pounds, which is about forty-one kilograms.E1.This is what Cathy did yesterday.2.She got up at six thirty and cooked breakfast for her family.3.At seven thirty, she drove her two children to school.4.I took about forty-five minutes to take them to school.F1.After driving them to school, she returned home.2.For the next hour she cleaned the house.3.Then she talked to a friend on the phone.4.They decided to meet at a shopping mall.G1.She arrived at the mall at eleven forty-five.2.She met her friend in a coffee shop.3.They ate lunch together and then they went shopping.4.There was a big sale at one of the stores.5.There were discounts of up to fifty percent so they bought clothes for theirkids.H1.At one thirty here friend wanted to go to a fitness center(健身中心).2.They usually worked out three times a week.3.But yesterday, Cathy had a headache.4.She decided not to work out.5.Instead, she decided to go home.I1.When she got home, her headache was worse.2.She took some medicine and had a short nap.3.After her nap, she felt a bit better.4.Her headache was gone.J1.Her kids came home at five o’clock.2.They were all hungry, but she did not feel like cooking.3.Instead, she decided to order some food.4.She ordered some Chinese food.K1.Forty-five later, the delivery man came.2.She paid him for the food and then they ate dinner.3.The food was good and not very expensive.4.It was a nice change.5.Her husband didn’t come home until late.6.That was good, because he doesn’t like Chinese food.练:1.She weights ninety pounds less than he does.2.The man on the top left weighs the most.3.The girl is very thin and light.4.The boy is twenty pounds heavier than the girl.5.After she paid for the food, they ate dinner.L2-U3-4/4 vocabularyDifferent voice/ different subject1.Here are some things to listen to.2.Most people enjoy listening to music.3.There are many types of music, including classical and jazz.4.Listening to the radio.5.We can listen to the news, music and conversation on the radio.6.Listening to the sounds of the city.7.The sounds of the city are often loud and noisy.8.Listening to the sounds of nature.9.Early in the morning, we can often hear the sound of birds singing.10.Listening to other people.11.Sometimes listening to others is interesting.B1.Here are some school subjects.2.In science class, we learn about our world.3.Science is where we learn about forces such as gravity(重力,地心引力).4.In the math class we learn about numbers and shapes.5.We learn how to count and how to find the area of different shapes.6.In history class, we learn about the pass.7.We learn about how our country began and how it has changed.8.In geography(地理学) class we learn about the land and people of our planetearth.9.We learn about oceans, mountains and rivers.10.In literature(文学) class, we read and study great books and poems.11.We learned to write down our ideas and we learn how others think and feel. 练:1.Some people are very interesting, so we enjoy listening to them.2.Some people talk too much so we want them to stop talking.3.If we don’t know math, we can’t do business or build things.L2-U3-4/4 dialogueL isa’s morning /make reservation with doctorA1.Lisa, don’t f orget to take your things to school today.2.I won’t mum, I’m sorry about yesterday.3.And remember to pay back your money your own.4.Do you have enough money?5.I won’t forget, and I have enough money.6.How much is that?7.I have fifteen dollars.B1.Do you need anything else?2.No, I don’t think so.3.Are you ready for your history test?4.Yes, I’m ready.5.I did all the homework.C1.What about your iPhone?2.Do you have it?3.Oh, thanks, it’s in my room.4.Could you get it for me?5.No, you have to get it for yourself.6.Then maybe yo u won’t forget it again.D1.Hey, I can’t find it.2.It’s not in my bed.3.could you call me?4.Ok, I’m calling you now, and it’s ringing.5.Do you hear it?6.No, I don’t. I don’t hear it.E1.Why don’t you look in the bathroom.2.Maybe you left there.3.Ok, I’ll look in th e bathroom.4.It’s getting late, so hurry or you’ll miss your bus.5.Yes, I hear it, it’s in the bathroom.6.I put it there when I brush my teeth.F1.Good, now, hopefully you have everything you need for today.2.Thanks mom.3.I think I have everything.4.Good luck with your test.5.I won’t be here when you come home this afternoon.6.Ok, have a good day.G1.G ood morning, doctor’s office.2.Hello, my little girl has a bad sore throat.3.We’d like to see the doctor as soon as possible.4.Can it wait until tomorrow?5.No, it can wait until tomorrow.6.If the doctor can’t see her today, could we see someone else?H1.Let me check his schedule.-----WORD格式--可编辑--专业资料-----2.Yes, he has time to eleven o’clock.3.She may have to wait.4.That’s Ok, we can wait.5.Ok, your last name, please.6.Jackson.7.Ok, Mrs. Jackson.8.Does your little girl have a fever?9.Yes, she does. It’s around thirty nine.I1.Please stay in bed and drink plenty of liquids.2.Some tea may help with the pain.3.Thank you, we’ll be there at eleven o’clock.--完整版学习资料分享----。
Technical Report TR2004-850Computer Science DepartmentCourant Institute of Mathematical SciencesNew York UniversityOpticalflow estimation as distributed optimization problem-an aVLSI implementationauthor:Alan A.StockerMember IEEEHoward Hughes Medical Institute,Center for Neural Science andCourant Institute of Mathematical SciencesNew York University4Washington Place Rm809New York,NY10003-1056U.S.A.phone:+1(212)9928752fax:+1(212)9954011email:alan.stocker@ehome:/∼alanAbstractI present a new focal-plane analog VLSI sensor that estimates opticalflow in twovisual dimensions.The chip significantly improves previous approaches both with re-spect to the applied model of opticalflow estimation as well as the actual hardwareimplementation.Its distributed computational architecture consists of an array oflocally connected motion units that collectively solve for the unique optimal opticalflow estimate.The novel gradient-based motion model assumes visual motion to betranslational,smooth and biased.The model guarantees that the estimation prob-lem is computationally well-posed regardless of the visual input.Model parameterscan be globally adjusted,leading to a rich output behavior.Varying the smoothnessstrength,for example,can provide a continuous spectrum of motion estimates,rang-ing from normal to global opticalflow.Unlike approaches that rely on the explicitmatching of brightness edges in space or time,the applied gradient-based model as-sures spatiotemporal continuity on visual information.The non-linear coupling of theindividual motion units improves the resulting opticalflow estimate because it re-duces spatial smoothing across large velocity differences.Extended measurements ofa30x30array prototype sensor under real-world conditions demonstrate the validityof the model and the robustness and functionality of the implementation.index:visual motion perception,2-D opticalflow,constraint optimization,gradient descent,aVLSI,analog network,collective computation,neuromorphic,feedback,non-linear smoothing,non-linear bias1MotivationThe ability to estimate motion using visual information is important for any natural and artificial agent behaving in a dynamical visual environment.Knowing the relative motions between different objects as well as between objects and the agent is crucial for a cognitive perception of the environment and thus a requisite for intelligent behavior.However,the demand for real-time processing and the limited resources available on freely behaving agents impose severe constraints that require an efficient computational systems in terms of processing speed,energy consumption and its physical dimensions.These requirements strongly favor analog VLSI(aVLSI)implementations of highly parallel and distributed computational architectures.In particular,such implementations become appealing when image sensing and motion computation can be combined within a single sensor.Ideally, such a sensor consists of a topographically uniform array of identical processing units, each providing a local estimate of visual motion at its particular image location.In suchabFigure1:The aperture problem.(a)Translational image motion induces locally ambiguous visual motion percepts.(b)Vector averaging of the normalflowfield(dashed arrow)does not lead to the correct global motion.Instead,only the intersection of the constraint lines provides the correct common object motion(bold solid arrow).architecture,processing power scales with array size,thus keeping the processing speed independent of spatial image resolution.Local visual motion is usually represented by a vectorfield,referred to as opticalflow.Analog VLSI circuits require significantly less power and silicon area than digital circuits for computational tasks of comparable complexity[1]. Furthermore,time-continuous analog processing matches the the continuous nature of visual motion information.Temporal aliasing artifacts do not occur while they,in contrast,can be a significant problem in clocked,sequential circuit implementations,in particular when limited to low frame-rates[2].Visual information is–in general–ambiguous and does not allow a unique local in-terpretation of visual motion,a major reason being the aperture problem[3].A priori assumptions are necessary in order to resolve the ambiguities.Such assumptions instan-tiate the motion model of the expected visual motion.Thus,the computational task and the resulting quality of the opticalflow estimate can vary substantially depending on the complexity of the chosen model.Figure1a illustrates the aperture problem for a simple mo-tion scene under noise-free conditions.Observations through apertures showing zero-order (aperture A)orfirst-order spatiotemporal brightness patterns(apertures B and C)do not allow an unambiguous local estimate of the visual motion.The presence of a higher-order brightness pattern as e.g.the corner of the triangle in aperture D,would allow to uniquely determine the local visual motion without further assumptions,however requiring complex spatiotemporalfilters to account for all possible patterns.Alternatively,instead of resolv-ing complex patterns locally,visual information could be spatially integrated,combining the perception through multiple,sufficiently small apertures such that the spatiotemporalpattern within each aperture can be well approximated to be maximally offirst -bining the constraints imposed by thefirst-order patterns then ideally leads to the unique and correct estimate of object motion.It is important to realize,that the vector average of the normalflow,that is the opticalflow being perpendicular to thefirst-order pattern ori-entation in each aperture,usually does not coincide with this collective constraint-solving estimate(see Figure1b).The problem remains to decide how the constraints imposed by the observations at the different apertures are weighted,i.e.which apertures observe a common visual motion source and should be combined.Dynamical grouping processes have been suggested that assign varying ensembles of apertures to different image objects at any time,leading to an individual estimate of visual motion for each ensemble(object) [4,5,6,7,8].Although the aVLSI implementation of an opticalflow sensor with dynam-ical grouping process has been reported recently[9],such dynamic processes will not be discussed further in this paper.Just note,that spatial integration is a necessary requisite to solve for the correct object motion.2Review of aVLSI visual motion sensorsMost of the known aVLSI motion sensors estimate visual motion only along a given single spatial orientation which significantly simplifies the computational problem.They can be classified in methods performing explicit matching in time-domain e.g.[10,11,12,13,14], gradient based methods e.g.[15,16]and implicity matching or correlation based methods, that follow insect vision e.g[17,18,19,20,21,22].Only a few2-D visual motion sensors have been reported.In the following,theses approaches are briefly reviewed according to their applied motion models.Afirst class es-timates the direction of normalflow.Higgins and colleagues[23]presented two focal-plane implementations that provide a quantized signal of the direction(8directions)of normal flow.The two architectures are functional very similar:the occurrence of a brightness edge at a single image location is detected and compared with its re-appearance at neighboring locations.Deutschmann and Koch[24]proposed an approach to estimate the direction of normalflow by multiplying the spatial and temporal derivatives of the brightness distribu-tion in the image.The limited linear range of the applied multiplier circuits impairs the correct directional estimate,leading to an increasing emphasis on diagonal visual motions for increasing stimulus contrasts.The output signal is monotonic in visual speed for a given stimulus contrast but substantially varies contrast.Another class of implementation allows to estimate direction and speed of normalflow. Jiang and Wu[25]reported a correlation based approach to estimate normalflow.Motion isreported if the time-of-travel of the extracted edges between neighbor motion units matches a pre-set delay time.Since the tuning is very narrow,the sensor is fairly limited and can-not report arbitrary visual velocities without being continuously re-tuned.More practical sensors were reported by Kramer et.al.[26],applying explicit matching in the time-domain. Two circuits are presented where the time-of-travel of an brightness edge in the image is measured either by eliciting a monotonic decaying function with arrival of an edge and sam-pling the functions value at the time the edge passes a neighboring pixel,or,by measuring the amount of overlap of twofixed-size pulses of neighboring pixels that each are triggered by the arrival of a brightness edge.A similar approach but different implementation was proposed by Etienne-Cummings et.al[27].Where as in[26]temporal intensity changes are assumed to represent brightness edges,here,brightness edges arefirst extracted in the spatial domain before matching was performed in the time-domain.This has the advantage that also very slow speeds can accurately be detected.None of the above approaches,however,perform spatial integration beyond averaging in order to solve the aperture problem.Tanner and Mead[28]reported a sensor with an array size of8x8pixels that provides two output voltages each representing the two components of the global motion vector.But,measured motion data was never explicitly shown and the sensor was reported to be fragile,even under well-controlled laboratory conditions.Nevertheless,it was thefirst hardware example of a collective computational approach to visual motion estimation.Subsequent attempts[29]failed to result in a more robust implementation.Only in a previous paper[30],we presented afirst improvement that also allowed estimation of smooth opticalflow besides a global motion estimate.The prototype implementation with a7x7array was functional although it was rather limited by its small linear output range.Another recent attempts[31]incorporated segmentation properties into our previous approach but were not able to demonstrate robust behavior under real-world conditions.The2-D opticalflow sensor reported here represents a significant improvement and further development of all previous approaches,both in terms of the applied motion model as well as its aVLSI implementation.It is thefirst robust and fully functional sensor of its kind.My recently documented motion segmentation chip[9]shares an identical circuit design for its motion units.3Optical Flow ModelFor analytical reasons,we define the input of the proposed model to be the spatiotemporalE(x,y,t),and E t=∂gradients E x=∂∂ybution E(x,y,t)in the image.We will later discuss the extraction of these spatiotemporal gradients in the actual focal-plane implementation.The model’s output is the opticalflow field v(x,y,t)=(u(x,y,t),v(x,y,t))that represents the instantaneous estimate of visual motion.To increase readability,space and time dependence will not be explicitly expressed in subsequent annotations although it is implicitly always present.The applied motion model assumes that the brightness constancy constraint[32] holds and that the opticalflow varies smoothly in space.Following Horn and Schunck[33], these two constraints can be formulated as optimization problem for which the desired opticalflow estimate is the optimal solution(see also[34]).However,it can easily be verified that using only these two constraints results in an ill-posed estimation problem for particular visual input ly,when only zero andfirst order brightness patterns of equal orientation are present throughout the whole image.It is highly desirable for a physical system such as e.g.an aVLSI sensor that the computational function implemented is always well-posed.Otherwise,the system’s behavior is unpredictable,driven by noise and non-idealities of the implementation.Therefore,the so-called bias constraint is added, expressed as the cost functionB(v)=(u−u ref)2+(v−v ref)2,(1)which weakly biases the opticalflow estimate to some predefined reference motion(u ref,v ref). The reference motion can be understood as the a priori expected motion in case the visual information content is unreliable or missing.In contrast to similar suggested formula-tions[35,36,37],the reference motion is not necessarily assumed to be zero.Much more, it could be adaptive to account for the statistics of its visual environment;the reference motion could e.g.represent the statistical mean of the experienced visual motion.However, such adaptation mechanisms will not be further discussed in this article.Combining the model of Horn and Schunck with the additional bias constraint(1),we now formulate the following constraint optimization problem over all nodes i,j in a discrete,orthogonal n×m image space:Given the input E xij ,E yijand E tijand a reference motionvref=(u ref,v ref),find the opticalflowfield v ij such that the cost functionH(v ij;ρ,σ)=ni=1m j=1 (E x ij u ij+E y ij v ij+E t ij)2+ρ((∆u ij)2+(∆v ij)2)+σ((u ij−u ref)2+(v ij−v ref)2) (2)is minimal1.The positive parametersσ>0andρ≥0determine the relative influence of each constraint.2h 2+x i,j+1−x i,j−1σV U ]x i= [1...n ]Figure 2:A single unit of the optical flow network.4Optical Flow NetworkThe optimization problem (2)is convex for any given input.Thus,a linear dynamical system that performs gradient descent on the cost function (2)is guaranteed to provide the optimal estimate once it reaches steady-state.Partial differentiation of (2)results in a system of 2n ×m linear partial differential equations˙u ij =−1CE y ij (E x ij u ij +E y ij v ij +E t ij )−ρ(v i +1,j +v i −1,j +v i,j +1+v i,j −1−4v ij )+σ(v ij −v ref )(3)with C being a positive constant.These first-order partial differential equations exactly describe the dynamics of two ac-tively coupled resistive networks.Figure 2illustrates a single unit of such coupled networks.Identifying the local estimate of optical flow with the voltages U ij ,V ij 2,the terms in straight brackets (3)represent the sum of currents that charges up or down the capacitances C until equilibrium is reached.Each constraint of the cost function (2)has a physical counterpart:smoothness is enforced by the resistive networks with lateral conductances ρ.The biasconstraint is implemented as the leak conductanceσto the reference motion represented by the potentials V ref and U ref.Only the implementation of the brightness constancy constraint needs some active cir-cuitry represented by the“constraint-boxes”[38]A and B that inject or sink the currentsF ui,j ∝−E xij(E xiju ij+E yijv ij+E tij)and F vi,j∝−E yij(E xiju ij+E yijv ij+E tij),(4)respectively.These correction currents represent the violation of the constraint and are computed in a cross-coupled feedback loops.We exploit the natural dynamics of a physical system to solve a visual perception problem[39]:The solution of the optimization problem is represented by the steady-state of the analog electronic network.The system is assumed to be in steady-state at any time.This approximately holds if the time-constant of the network is negligible compared to the dynamics of the input.The appropriate control of the node capacitance C and the current levels in the implementation can ensure a close-to-optimal solution for reasonably slow input dynamics.Note,that the dynamics of the network strongly depend on the spatiotemporal energy of the visual input.The characteristics of the model and therefore the computational behavior of the optical flow network are determined by the relative weight of the three constraints,determined by the lateral and vertical network conductancesρandσ,respectively.According to the strength of these conductances,the network accounts for different models of visual motion estimation such as normalflow,smooth opticalflow or globalflow.The presented optical flow sensor allows to globally adjustρandσwhich is certainly one of its advantages.5Circuit Architecture and ImplementationThe complete circuit schematics of a single motion unit of the opticalflow sensor is shown in Figure3.Since the input(spatiotemporal intensity gradients)and the output(the compo-nents of the local opticalflow vector)can take on positive and negative values,a differential encoding of the variables is applied consistently throughout the circuit:The variables are encoded as the difference of two voltages(or currents).However,referencing the voltages U+and V+to afixed null potential V0substantially reduces the implementation complexity by requiring only two single line resistive networks.Note,that the reference motion for the bias constraint is chosen to be zero,thus U ref,V ref=V0.5.1Extraction of the brightness gradientsThe circuit schematics contains all building blocks of a single unit.At afirst processing stage,the spatiotemporal brightness gradients have to be estimated.Each pixel includes a logarithmic,adaptive photoreceptor[40]with adjustable adaptation rate[41].The temporal gradient is extracted by a hysteretic differentiator circuit[42]providing the currents E t−,E t+that represent the rectified temporal derivative.The adjustable source potentials HD tweak+,HD tweak-allow to control the output current gain of the differentiator.The spatial derivatives E x and E y are estimated asfirst-order approximation of the brightness gradients,thus the central difference value of the photoreceptor output of nearest neighbors(P h x+−P h x−)and(P h y+−P h y−)respectively(see schematics in Figure3).While continuous-time temporal differentiation avoids any temporal sampling artifacts,discrete spatial sampling of the image brightness significantly affects the visual motion estimate.Let us consider a simplified,one-dimensional arrangement as depicted in Figure4a, where a sine-wave grating moving with afixed velocity v is presented.We describe the brightness distribution in the focal-plane as E(x,t)=sin(kx−ωt)with x,t∈R,where ω=kv and k is the spatial frequency of the projected stimulus.First order approxi-mation of the spatial gradient at location x i thereby reduces to the difference operator ∆x i=(x i+1−x i−1)/2d,where d is the sampling distance(pixel size).Assuming a non-zero sampling size D(photodiode),the discrete spatial gradient becomesE x(x i,t)=1D x i+d+D/2x i+d−D/2sin(k(ξ−vt))dξ−1kDdsin(kd)sin(kD/2)cos(k(x i−vt)).(5)Similarly,the temporal gradient at location x i is found to beE t(x i,t)=∂D x i+D/2x i−D/2sin(k(ξ−vt))dξ=−2vσ+E x(x i,t)2(7)Substitution of(6)and(5)into(7)and assuming V ref=0,wefindV out(x i,t)=v kdγ+cos(kx i−ωt)2withγ=σk2d2D2ak=0.5k=1k=1.5xi-1xixi+1bπspatial frequency k [/d]motionresponse[a.u.]v0.030.112Figure4:Spatial sampling and its effect on the motion response.(a)Spatial sampling of sinusoidal brightness patterns of different spatial frequencies k(given in units of the Nyquist frequency[π/d])in one visual dimension.(b)The expected response of the optical flow sensor according to(7)as a function of the spatial frequency k of a sinewave stimulus moving with velocity v.The dashed curve is the time-averaged response(σ=0.05,δ=1/3).We now can predict the motion response of a single isolated unit as a function of spatial frequency of the stimulus for different weights of the bias constraintσas well as different fill-factorsδ=D/d.Figure4b shows the predicted peak output response for different parameter values as a function of the spatial stimulus frequency k,given in units of the Nyquist frequency[π/d].Note that the response strongly depends on k.Only for low spatial frequencies,the motion output well approximates the correct velocity.For very low frequencies,the local stimulus contrast diminishes and the non-zeroσbiases the output response toward the reference motion V0=0.As the spatial frequency approaches the Nyquist frequency,the response increases to unrealistically high values and changes sign at k∈Z+[π/d].Increasing thefill factor reduces the effect,although only marginally for reasonable values ofδand smallσ.Clearly,if the bias constraint is not imposed,thusσ=0, Equation(8)reduces toV out=vkdthe equation which means that the spatial low-passfiltering of the photodiode only affects the motion estimate if the bias constraint is imposed.Secondly,the output(9)does not depend on space nor time.Forσ>0,however,γis always non-zero and the motion response becomes phase-dependent due to the remaining frequency terms in(8).As a consequence,the time-averaged motion response over the duration of a complete stimulus cycle is always less than the peak response as shown in Figure4b.Phase-dependence is a direct consequence of the bias constraint.However,spatial integration of visual information due to collective computation in the opticalflow network allows to partially overcome the phase-dependent response of a single unit.Note,that thefirst-order difference approximation of the spatial brightness gradients in two visual dimensions is not rotationally invariant.Unfortunately,more elaborate gradient approximations[43]are not feasible for a compact focal-plane implementation.5.2Wide-linear range multiplierThe design of the multiplier circuit needed to compute E x u and E y v respectively,is cru-cial for a successful aVLSI implementation of the opticalflow network.Being part of the recurrent feedback loops that generate the error-correction signals F u and F v,the demands on the multiplier circuit are high.It has to operate in four quadrants,providing a wide linear output-range with respect to each multiplicand.Offsets should be minimal because they directly impose offsets in the opticalflow estimate.Furthermore,the design needs to be compact in order to allow a small pixel size.The original Gilbert multiplier[44]meets these requirements fairly well,with the exception of its small linear range when operated in sub-threshold.In above-threshold operation,however,its linear range is significantly increased due to a transconductance change of the individual differential pairs[45].The multiplier circuit proposed here,shown in detail in Figure5a,embeds a Gilbert multiplier in an additional outer differential pair.The idea is to operate the Gilbert mul-tiplier above-threshold to increase the linear range but in addition,to rescale the output currents to sub-threshold level such that the current levels in the feedback loop match.The scaling is approximately linear,thusI b2I out=I out+−I out−≈3Shockley equation:I diode=I0(exp(1)−1).kTaI out-I out+b0.20.40.60.811010101010-12-10-8-6-4emittercurrentIE[A]base-emitter voltage VBE[V]10-2c-1-0.500.51-4-224x 10-9∆VAdifferential input [V]Iout[A]d-1-0.500.51-4-224x 10-9differential input ∆V B [V]Iout[A]Figure5:Wide linear-range multiplier.(a)Circuit schematics.(b)The measured emitter currents for a native npn-bipolar transistor and the vertical pnp-bipolar of a pFET.(c) Measured output currents of the wide linear-range multiplier as a function of the applied input voltage at the lower differential pair∆V A=V2−V1.Each curve represents the output current for afixed differential voltage∆V B=V4−V3=[0,0.05,0.1,0.15,0.2,0.3,0.4,0.5, 0.75V].(d)Same as c but now sweeping the upper differential pair input∆V B.ideal(n=1)behavior[46]which has been verified by the measurements shown in Figure5b. Also,the voltage drop across the diodes is such that the gate voltage of the outer nFETs are typically within one volt below V dd,meaning that their gate-bulk potentials are large. At such levels,κasymptotically approaches unity because the capacitance of the depletion layer becomes negligibly small compared to the gate-oxide capacitance.Thus,κn≈1andwe can safely assume a linear scaling of the multiplier output current.Base-emitter junctions can be exploited either using native bipolar transistors in a gen-uine BiCMOS process or the vertical bipolar transistors in standard CMOS technology.In practice,however,it is necessary to use a native BiCMOS process in order to implement the multiplier circuit correctly:Figure5b displays the measured base-emitter voltages V BE as a function of the applied emitter current I E for both,a vertical pnp-bipolar transistor in a typical p-substrate CMOS process and a native npn-bipolar transistor in a genuine BiC-MOS process.At current levels above1µA,however,the vertical bipolar starts to deviate significantly from the desired exponential characteristics due to high-level injection caused by the relative light doping of the base(well)[46].This occurs already at a current level that is significantly below the range where the multiplier core is preferably operated at. The exponential regime of the native bipolar,however,extends up to0.1mA.Although a CMOS implementation of the diodes is desirable to avoid the more complex and expensive BiCMOS process,it would severely impair the linear scaling(10).Figure5c and d show the measured output of the multiplier for sweeping either of the two input voltages.For the applied bias voltages,the linear range is approximately±0.5 V and slightly smaller for the upper differential input.Note,that the measurements were obtained from a test circuit with identical layout to the one within each pixel.Offsets are small.The circuit is compact and allows to control the linear range and the output-current level independently by the two bias voltages V b1and V b2.The disadvantages are the increased power consumption caused by the above-threshold operation and the need for BiCMOS technology.5.3Output conductance of the feedback-loopConsider the current equilibrium at one of the capacitive nodes in Figure2.In steady state, we assume all currents onto this node to represent the deviation to the different constraints according to(3).To be completely true,this would require the feedback current to be generated by ideal current sources A and B.However,the small but present output con-ductance of the feedback node causes some extra current that shifts the current equilibrium. This output conductance can be understood as imposing a second bias constraint on the estimation problem.It biases the capacitive node to a reference voltage that is intrinsic and depends on various parameters like the strength of the feedback current or the Early voltages of the transistors.In general,this reference voltage is not identical with V0.Thus, the superposition of the two bias currents has an asymmetric influence on thefinal motion estimate.Since the total correction currents are typically weak,the effect is significant.The aim,therefore,is to reduce the output conductances of the feedback loop as muchas possible.The applied cascode current mirror circuit reveals a substantial decrease in output conductance compared with a simple current mirror used in previous related imple-mentations[28,29,30].Neglecting any junction leakages in drain and source,wefind the total output conductance at the individual capacitive nodes to beg o=g oN+g oP=F+·U TV E,P1V E,P2,(11)where F+−F−is the total feedback current,U T the thermal voltage and V E,X the Early voltages of the different transistors in the cascode current mirror(see box in Figure3).5.4Effect of Non-LinearitiesGradient descent on the cost function(2)defines a system of coupled,but linear partial differential equations.However,a linear translation into silicon is hardly possible.In the following,some important non-linearities of the implementation and their effect on the opticalflow estimate are discussed.5.4.1Saturation in the feedback-loopThefirst important non-linearity is caused by the saturation of the multiplier circuit in the feedback loop.As shown in Figure5,the output current saturates for larger input voltages. What does this mean in terms of the expected motion output of a single motion unit?For the sake of illustration,we once more assume a one-dimensional array of the optical flow network.Furthermore,we disable all lateral connections and neglect the bias constraint (ρ=0,σ=0).Then,for given non-zero brightness gradients,the circuit ideally satisfies the brightness constancy constraintE x u+E t=0.(12) Now,assume that the multiplication E x u is replaced by f(u)|Exwith E x constant,where f describes the saturating output characteristics of the proposed multiplier circuit.For reasonsof simplicity we assume the output of the multiplier core I coreoutto follow a simple sigmoidal function tanh(u)which is only qualitatively correct[45].We can rewrite the multiplieroutput(10)as f(u)|Ex =c ExI b2/I b1tanh(u),where the constant c Exis proportional to agiven E x.Since f(u)is one to one,we can solve(12)for the motion response andfindu=f−1(−E t)=−artanh(I b1。
is-is路由协议知识点IS-IS(Intermediate System to Intermediate System)是一种内部网关协议(IGP),用于在计算机网络中进行路由协议。
IS-IS最初是ISO(国际标准化组织)制定的一种协议,用于在OSI (开放系统互联)参考模型中的中间系统之间进行路由。
后来,IS-IS被广泛应用于互联网和大型企业网络中。
IS-IS协议提供了以下几个关键的功能和特点:1. Hierarchy(分层结构):IS-IS使用等级结构,将网络划分为区域,每个区域有一个区域边界路由器(Level 1 router)来处理区域内的路由信息。
多个区域构成了一个域(domain),有一个域边界路由器(Level 2 router)来处理域内的路由信息。
这种分层结构使得IS-IS在大型网络中具有高效的可扩展性。
2. 基于SPF算法的路由计算:IS-IS使用单一最短路径优先(SPF)算法来计算最佳路径。
该算法根据链路权重来选择最短路径,以确保数据能够通过最短路径快速到达目的地。
3. 域间路由:IS-IS支持域之间的路由,即Level 2路由。
通过域间路由,不同区域的网络可以相互通信和交换路由信息。
域间路由信息是通过域边界路由器相互交换的。
IS-IS使用了一个称为“TN(transport network)”的协议来传递域间路由信息。
4. 基于链路状态的路由更新:IS-IS使用链路状态数据库(Link State Database)来记录网络拓扑信息。
每个IS-IS节点都会维护一个完整的链路状态数据库,并将其与邻居节点共享。
链路状态数据库中包含了网络中所有链路的状态和代价,用于计算最短路径和进行路由更新。
5. 支持多种网络层协议:IS-IS不仅可以运行在IPv4网络中,也可以运行在IPv6网络中。
IS-IS通过将网络层协议与其上面的数据链路层协议(如Ethernet、Frame Relay等)分离,实现了对不同网络层协议的支持。
is-is 原理IS-IS(Intermediate System to Intermediate System)是一种基于链路状态的路由协议,用于在计算机网络中交换路由信息。
它是一种开放标准的协议,最初由ISO(国际标准化组织)定义,后来被引入到TCP/IP协议族中。
IS-IS协议的基本原理是通过交换链路状态数据包(LSP)来建立和维护一个网络拓扑图,以确定最佳的路由路径。
在IS-IS网络中,每个中间系统(Intermediate System,简称IS)都会维护一个链路状态数据库(Link State Database),记录了与其相邻的IS之间的链路状态信息。
IS-IS协议使用了SPF(Shortest Path First)算法,通过计算每个IS到其他IS之间的最短路径,确定最佳的路由路径。
SPF算法基于Dijkstra算法,通过计算每个节点到其他节点的最短路径代价,选择路径代价最小的路径作为最佳路径。
IS-IS协议具有以下特点:1. 分层结构:IS-IS协议将网络划分为不同的区域,每个区域内部使用IS-IS协议进行路由,而区域之间则通过边界路由器进行通信。
这种分层结构可以提高网络的可扩展性和灵活性。
2. 支持IPv4和IPv6:IS-IS协议既可以支持IPv4网络,也可以支持IPv6网络。
对于IPv4网络,IS-IS使用TLV(Type-Length-Value)格式来传输路由信息;对于IPv6网络,IS-IS使用扩展格式来传输路由信息。
3. 冗余容错:IS-IS协议支持多路径冗余,当某条路径出现故障时,可以快速切换到备用路径,保证网络的可靠性和容错性。
4. 可扩展性:IS-IS协议可以灵活地适应不同规模的网络,可以根据网络的需求进行区域划分和路由聚合,提高网络的可扩展性。
5. 快速收敛:IS-IS协议通过快速地传播链路状态信息,可以在网络发生拓扑变化时快速收敛,减少网络的震荡和不稳定性。