Homework 4 Question 4 (Latency) - Incorrect grading?

During the lecture unit, professor clearly states that if we could replace fiber optics with vacuum, we could accomplish the a great speed increase. I realize that in the current state of technology this is not currently possible (and might never be possible in the future) but the way homework question was phrased, we cannot assume that.

I believe the correct answer to the question should be "Replacing all the fiber optics with vacuum".

asked 20 Mar '12, 20:11

Mike%20Zaloznyy's gravatar image

Mike Zaloznyy
1.0k2866
accept rate: 40%

And I agree with you Mike. He made the point in the lecture/demo and it was a good one. And, that on top of the parenthetical note on the question, which seemed to be stressing the huge improvement.

(20 Mar '12, 20:14) Clair Dunn Clair%20Dunn's gravatar image
5

Hmmm, I heard David say that if it were one vacuum - the question was to replace ALL fiber optics with vacuums - and that leaves still the 20 routers that have to handle the transfers. The in-between travel speed would be tremendous, but you still have to wait around while the router does its thing.

(20 Mar '12, 20:35) Marlen Waaijer Marlen%20Waaijer's gravatar image
5

@clair He clearly states in the video that most of the latency is due to delays at routing points. The point is that the decrease in latency achieved by replacing the cables with vacuums would be negligible since the longest delays would still occur at routing points.

(20 Mar '12, 20:36) curiousborg curiousborg's gravatar image
4

I fell for it too. My brain told me the answer was decreasing the routing points, but for some reason I took the question as a "trick" question and chose the vacuum answer because of what I thought I heard during the lecture. Got an 86% homework grade because of it :-) Listening to the video explanation, however, reducing the routing points is correct.

(20 Mar '12, 20:39) Joel Marcey-1 Joel%20Marcey-1's gravatar image

Yep - I fell for it also - Kicking myself now :/

(20 Mar '12, 21:14) Ken Jepson-1 Ken%20Jepson-1's gravatar image

@Joel. The same thing happened to me -- including the 86% grade. What still annoys me is that I think I would have gotten it right had he not unhelpfully qualified the statement of the question with his parenthetical remark: "Note: several of them would improve the latency a little bit, but only the one answer which would produce the biggest improvement is correct." I remembered him making such a big to-do about how dramatically faster vacuum tubes are that I didn't think they would only make a "little" improvement in the latency. So I picked that answer. However, when I went back to review his exact wording (see the Traveling Data question from Unit 4), I discovered that the point you and I both thought he had made (vacuum-tube connections are super fast) was very different from the point he had actually made (direct vacuum-tube connections are super fast; in fact, direct fiber-optic connections are super fast, too). I've been really happy with the course, so I hate to quibble, but I think the fact that so many people made that mistake means that he could have been a bit clearer in emphasizing that the true limiting factors are the routers, not the wires. Of course, having gotten the question wrong and really thought about why, I probably learned the material better than I otherwise would have!

(20 Mar '12, 21:14) George Potter-1 George%20Potter-1's gravatar image

I got the answer right but unfortunetly I clicked the wrong one when submitting the answer :\ I also got 86% because of that :\

(21 Mar '12, 06:21) Jan Hančič Jan%20Han%C4%8Di%C4%8D's gravatar image

10 Answers:

The decrease in time for fibre optics -> vacuum is very small compared with the time it takes for the packets to get through the routers.

For the 100 ms journey in the videos, approximately 72 ms is taken at routers and 28 ms 'on the wire' since the speed of light along optic fibre is about half what it is in a vacuum.. There would be a saving of approximately 14 ms if the fibre optics were replaced by vacuums. If the number of routers is cut to 4, then that's reducing the time spend at routers to 1/5 of the original 72 seconds, which is around 14 ms. That saves around 58 ms in contrast to 14 ms for the vacuums.

link

answered 20 Mar '12, 20:20

fnenu-1's gravatar image

fnenu-1 ♦♦
18.5k1981231

edited 20 Mar '12, 20:21

No one was mentioning the time for the packets going through. Assume 0 and it will depend on the line only.

(20 Mar '12, 20:24) TooMuchIdeas TooMuchIdeas's gravatar image
1

@TooMuchIdeas The whole point is that most of the latency is due to delays at routing points, he clearly states this in the video.

(20 Mar '12, 20:34) curiousborg curiousborg's gravatar image
2

The answer totally depends on the speed of the routers, which he is not mentioning in the video.

Take fnenu's example but just other numbers:

For the 100 ms journey in the videos, approximately 10 ms is taken at routers and 28 ms 'on the wire' since the speed of light along optic fibre is about half what it is in a vacuum.. There would be a saving of approximately 14 ms if the fibre optics were replaced by vacuums. If the number of routers is cut to 4, then that's reducing the time spend at routers to 1/5 of the original 10 seconds, which is around 2 ms. That saves around 8 ms in contrast to 14 ms for the vacuums.

(20 Mar '12, 21:13) TooMuchIdeas TooMuchIdeas's gravatar image
1

In Unit 4.18 (after the quiz), he states that most of the latency is not on the wires, but at routing points.

(20 Mar '12, 21:14) curiousborg curiousborg's gravatar image

Yes, but that was not the question. It was about reducing 20 hops to 4 without giving any numbers for those hops. Do the math yourself. I just knew nothing about the speed savings for reducing those 20 hops to 4. The math I was giving to you is not lying. As fnenu was giving an example where the answer is correct I was giving one where it is wrong.

(20 Mar '12, 21:21) TooMuchIdeas TooMuchIdeas's gravatar image

I asked a similar question, maybe my solution can help you out: http://www.udacity-forums.com/cs101/questions/43766/latency-question

(20 Mar '12, 21:24) jeremy low jeremy%20low's gravatar image

Yes but your numbers are not realistic, that example would never occur. I don't mean to be arguing back and forth with you pointlessly. The whole point is that the information processing done at routing points is always going to be relatively slowly compared to transmitting the information from point to that, if you didn't understand that before, hopefully you do now.

(20 Mar '12, 21:27) curiousborg curiousborg's gravatar image

Of course they are. Just read the post of XXnerd:
This ACM article says, "Modern routers such as the Cisco CRS-1 exhibit average latencies of about 100 microseconds."

http://queue.acm.org/detail.cfm?id=1530063

(20 Mar '12, 21:29) TooMuchIdeas TooMuchIdeas's gravatar image

@TooMuchIdeas, yes, 100 ms, you were suggesting 10ms.

(20 Mar '12, 21:31) curiousborg curiousborg's gravatar image

@curiousborg
100 microseconds. NOT 100 milliseconds.

100 microseconds are 0.1 milliseconds.
So 40 routers would only have 4 ms.

(20 Mar '12, 21:32) TooMuchIdeas TooMuchIdeas's gravatar image

I haven't seen that article before, I'll read it. I'll be very surprised if it concludes that fibre optic cables will ever be the bottleneck on any network.

Edit: I don't know enough about the hardware to really comment on this, but there are other considerations including network load that will affect the latency. I'm pretty sure though that it's safe to say, as in the answer video, that networks are not limited by the speed of light but by the routing points.

(20 Mar '12, 21:37) curiousborg curiousborg's gravatar image

May you take the words from CISCO:

"Latency with a hardware-assisted switch will be in the 4-to-20 microsecond range. The most reasonable processing delay that you can expect in practice should be 25 microseconds per hop. The processing delay on a software-based router can be considerably higher."

http://www.cisco.com/application/pdf/en/us/guest/netsol/ns407/c654/ccmigration_09186a008091d542.pdf

or may you should also read this article:
". Our Philadelphia–Palo Alto example would include approximately 30 of them in the round-trip path, making the total switching time latency about 3ms."

http://www86.homepage.villanova.edu/christopher.salembier/documents/physics.pdf

(20 Mar '12, 21:47) TooMuchIdeas TooMuchIdeas's gravatar image

I'm inclined to think that Dave and Peter know what they are talking about, but nevertheless I'll certainly ask this question in the next office hours.

(20 Mar '12, 23:14) curiousborg curiousborg's gravatar image

"For the 100 ms journey in the videos, approximately 10 ms is taken at routers and 28 ms 'on the wire'"

Where did you put the other 62 ms?

(21 Mar '12, 10:24) Anton Golov ♦ Anton%20Golov's gravatar image

It was an example for showing that routers latencies are not in the range of milliseconds, but microseconds.

(21 Mar '12, 19:49) TooMuchIdeas TooMuchIdeas's gravatar image
showing 10 of 15 show 5 more comments

The answer to this question can be found on Unit 4 Lecture 12 (Latency). If you listen carefully to what David says, he stated that,

If you could increase the speed of light, well that would make the network faster. Not a very realistic option and it wouldn't make it much faster because other than transit time even with the regular speed of light, the time it takes for the light to travel between the points is infinitesimally small compared to the time for all the other things like starting fires.

link

answered 20 Mar '12, 20:36

pmigueld's gravatar image

pmigueld
10616

I think this question should be thrown out. I got it "right" based on the example in the video, but the "right" answer assumes a few things that aren't right:

  1. It doesn't take 100 ms to get to Cambridge, MA from Palo Alto, CA. Traceroute shows round-trip time, so it actually takes 50 ms.
  2. They only did one test. There were a lot of routers, but the routers also seem pretty congested. Are the results always that slow? I notice that when I traceroute to MIT from here (Pacific Northwest), I get only 40 ms (one way). I notice that I go to Seattle to Chicago to Boston, with no hopping around from router to router in the congested New York area.
  3. The latency through routers is actually very quick, measured in microseconds, not milliseconds. I think our prof is assuming a much longer latency.
  4. The results of traceroute include any processing latency at the end hosts too, at least at the sender in our case (the destination host wasn't reached, so it's not relevant.)

Here are some documents about latency in routers:

This one compares IPv4 versus IPv6, but is still relevant. Note that the latency numbers for somewhat high-end routers (covered on the last page) is 28 microseconds. Of course this is a vendor measuring it, and they admit that the routers have pretty basic configs, but still, I bet that's orders of magnitude less than what our professor is assuming: http://www.cisco.com/web/strategy/docs/gov/IPv6perf_wp1f.pdf

This one compares Cisco and Juniper, supposedly written by an impartial 3rd party. Notice the 16 microsecond latency in the chart on page 4. http://www.bradreese.com/core-router-comparison-cisco-juniper.pdf

This ACM article says, "Modern routers such as the Cisco CRS-1 exhibit average latencies of about 100 microseconds." http://queue.acm.org/detail.cfm?id=1530063

So, I don't think the answer is as clear-cut as our professor assumes. Reducing the number of routers in a lot of cases won't help? What do others think?

link

answered 20 Mar '12, 21:12

XXnerd's gravatar image

XXnerd
92431428

edited 21 Mar '12, 10:25

Anton%20Golov's gravatar image

Anton Golov ♦
13.3k2174174

2

I totally agree with you. It totally depends on the speed of the routers. If it falls under a given threshold cutting down the 20 hops to 4 it wouldn't be as much saving as a vacuum line.

"Latency with a hardware-assisted switch will be in the 4-to-20 microsecond range. The most reasonable processing delay that you can expect in practice should be 25 microseconds per hop. The processing delay on a software-based router can be considerably higher."

http://www.cisco.com/application/pdf/en/us/guest/netsol/ns407/c654/ccmigration_09186a008091d542.pdf

(20 Mar '12, 21:27) TooMuchIdeas TooMuchIdeas's gravatar image

The speed of transmission will largely be based on the number of routers along the way, regardless of how they are interconnected.

link

answered 20 Mar '12, 20:16

Renzo's gravatar image

Renzo
313

edited 20 Mar '12, 20:16

Correct, those connection points are slowing the whole thing down much more.

(20 Mar '12, 20:18) Tom Vandenbosch Tom%20Vandenbosch's gravatar image

Not really. Just assume the time a station has to work on a packet is 0, the number of stations do not matter at all!

The answer was dealing with vacuum lines, so we have to also consider stations with a transmission time of 0.

Additionally it was not ask to work in ALL cases.

(20 Mar '12, 20:20) TooMuchIdeas TooMuchIdeas's gravatar image

If you put just zero it doesnt make any sense. I can also say if the time the data needs per mile in Fibre optic is 0, it doesnt change anything if you change the wires.
Got the Point?

(20 Mar '12, 20:26) Sebastian Po... Sebastian%20Pook-1's gravatar image

I got your point but it doesn't even make sense about asking If the line is a vacuum as it is not reality, but was still asked for. Got MY point?
If the answer is about vacuum lines, I can also assume about 0 latency routers. It was NOT specified.

Instead of 0 just use lightspeed routers. In this case it also wouldn't matter (point to point) distance.

(20 Mar '12, 20:28) TooMuchIdeas TooMuchIdeas's gravatar image

@TooMuchIdeas You can't assume zero time at a station, since that's actually where most of the time gets lost!

(20 Mar '12, 20:28) Tom Vandenbosch Tom%20Vandenbosch's gravatar image

It could be reality its just very expensive. And if you JUST assume zero it JUST doesnt make any sense. Try 0.001 as a number and you could make real calculations.

(20 Mar '12, 20:31) Sebastian Po... Sebastian%20Pook-1's gravatar image

As there were no constraints regarding the speed of the stations I can assume it. I just want to point out, that the questions is ambiguous cause of the lack of those constraints. No constraints - All assumptions are possible.

F.ex. The bandwidth question. If I assume there is a compression algorithm, which compresses all files to zero I can have 0 bandwidth ;-)

And this is why there has to be constraints for such questions.

(20 Mar '12, 20:43) TooMuchIdeas TooMuchIdeas's gravatar image

@TooMuchIdeas No, I'm afraid that assumption does not make any sense.

If you paid close attention to the video after the quiz in Unit4.18, the "constraints" you mention would have been clear to you.

(20 Mar '12, 20:59) curiousborg curiousborg's gravatar image

@curiousborg

Then just assume other numbers as in the example above: 10ms for the routers.
It depends on the time spent in the routers. But if the time spent in those 4 routers is bigger than the vacuum savings the answer is just wrong.

(20 Mar '12, 21:14) TooMuchIdeas TooMuchIdeas's gravatar image

@TooMuchIdeas The time spent passing through the routers will always be longer than the time spent on the wires. We're talking about the real world here, where the information processing that takes place at routing points takes a lot longer than just sending bits from one point to another.

The argument that we could assume that the routers work faster than the fiber optic cables is just silly.

(20 Mar '12, 21:18) curiousborg curiousborg's gravatar image
1

@couriousborg It's not a silly argument and it's not true that the time spent in routers "will always be longer." Do the math for high speed carrier class routers (with microsecond latency). Also keep in mind that traceroute isn't a good test. Routers put ICMP packets in a low priotiy queue. In addition, retuning an ICMP TTL Exceeded message is more "expensive" (higher latency) than simply forwarding an IP packet. The information processing is minimal and the routers are built to optimize this task. In another comment you said you would bring this up in office hours. Thanks for doing that.

(21 Mar '12, 00:53) XXnerd XXnerd's gravatar image

@curiousborg
cut through switching does exactly what you said: sending bits from one point to another as soon as the destination address is processed (in contrast to store and forward). This also reduces latency a lot.

Yes, we ARE talking about the real world here. And in the real world switching latency per hop is about 25 microseconds.

I never said routers work faster than the fibers. I just said they are working in the range of microseconds and not milliseconds.

(21 Mar '12, 19:54) TooMuchIdeas TooMuchIdeas's gravatar image
showing 10 of 12 show 2 more comments

Have you looked at the answer video?

link

answered 20 Mar '12, 20:31

PeterUdacity's gravatar image

PeterUdacity ♦♦
35.9k73219331

Going to check it out now.

(20 Mar '12, 23:04) Mike Zaloznyy Mike%20Zaloznyy's gravatar image

Peter -- I just learned late yesterday that there ARE answer videos/solutions - I can't wait to check them out from the beginning. And, Peter, I realize now why my answer was wrong, but the thing that gets me, is I had checked the right one as I was reading it, and changed it when I saw the vacuum answer, thinking this was a sort of trick/funny qeustion. Sigh.

(21 Mar '12, 03:52) Clair Dunn Clair%20Dunn's gravatar image

I do not remember that professor Dave mentioned cisco routers.
I just write down some things that Professor Dave told us in the answer video for travelling data about latency and I quote:

That speed of ligth trough optical fiber is about 50% slower than ligth travels in a vacuum.

If you have a VACUUM between Palo Alto and Cambridge, and no ROUTERS and nothing else along the way well then you could get your packets across about 7 times faster than the Internet actually does.

Then I do not understand the answer that PeterUdacity chose.

link

answered 21 Mar '12, 02:23

Edgar%20Lara's gravatar image

Edgar Lara
2

I also perceived it as such hence got the question wrong. But I guess the understanding is that you only replace the fibre optic with the vacuum, hence remain with the routers to which the fibre cables were attached. But I still have reservations with other aspects e.g. if the routers are actually quite fast.

(21 Mar '12, 09:47) gmirt gmirt's gravatar image

My figures (rounded) where slightly differents, but the result is exactly the same. From the lectures you know that 75% of the travel time is lost in routers. You know (lectures) that vaccum speed doubles fiber optic. For a travel time of 100ms we have 75 ms in routers and 25 ms travelling.

b) we still have 75 ms in the 20 routers + 25 / 2 ms replacing fiber optic = 87,5 ms

c) we still have 25 ms of travelling + 75 ms / 5 reducing routers by 1/5 = 40 ms

Then solution is c.

link

answered 21 Mar '12, 04:06

macjohn's gravatar image

macjohn
4.3k93372

edited 21 Mar '12, 04:27

Now I got it, thanks a lot, I was wrong

link

answered 21 Mar '12, 04:25

Edgar%20Lara's gravatar image

Edgar Lara
2

I think the distance between router to router matters. For example, in the lecture video, when 'tracerouting' on locations far apart, there were clear differences when one moved from a hop within the region to a hop outside that region. Wouldn't the replacement with vacuum, for instance between a router in the US and one in Antananarivo(madagascar) result into better transmission than lowering the number of routers?

link

answered 21 Mar '12, 10:11

gmirt's gravatar image

gmirt
123

It really has to do with the interpretation of the answer. I thought it was a single vacuum between the 2 points which would clearly be faster than 4 routers.

link

answered 30 May '12, 16:04

idoun's gravatar image

idoun
11257

Your answer
Question text:

Markdown Basics

  • *italic* or _italic_
  • **bold** or __bold__
  • link:[text](http://url.com/ "Title")
  • image?![alt text](/path/img.jpg "Title")
  • numbered list: 1. Foo 2. Bar
  • to add a line break simply add two spaces to where you would like the new line to be.
  • basic HTML tags are also supported

Tags:

×15,260
×515
×14
×11
×8

Asked: 20 Mar '12, 20:11

Seen: 921 times

Last updated: 30 May '12, 16:04