Revision Help - Networks!
Discussion
Hey all,
Can any one help me with the following questions?
I have no idea how you work this out but am I right in saying that I would need to take the median as the transfer rate would slow the further it gets from the sender?
As I don't know how to work out question A I can't answer this one yet. Again am I right in thinking that I should talk about Buffer management (Sliding windows), ARQ etc etc
Can any one help me with the following questions?
Question A said:
A company has offices some 3000km apart, and has taken up a Telephone Company offering a 10Gbps service. Calculate, with explanation, the maximum transfer rates for data links using TCP/IP over this service. You may assume that
a) the link is fibre-optic
b) the length of fibre between offices is exactly 1500km
c) The speed of light in fibre is 300,000km per second.
I have no idea how you work this out but am I right in saying that I would need to take the median as the transfer rate would slow the further it gets from the sender?
Question B said:
b) Two workstations, A&B, are simultaneoisly retrieving a 125MegaByte (MB) file from Server C over the link given in part (a), using TCP/IP. calculate with explanation, the shortest possible time for this retrieval for workstation A, and for workstation B.
As I don't know how to work out question A I can't answer this one yet. Again am I right in thinking that I should talk about Buffer management (Sliding windows), ARQ etc etc
What sort of level is this? A level, Undergrad, Postgrad or *shudder* industry certification such as MCSE, Cisco etc.?
I ask because there are all sorts of issues here. For instance, if they are offering a 10Gbit service then it doesn't matter what the distance or medium is, the rate is 10Gbit. As you say though, TCP/IP throughput depends on datagram size/overhead, packet size/overhead, error detection / correction, line noise, permittable latency, the amount of data being moved (and so on and so on...) many many variables.
If it's undergrad computing, my guess is that most of the answer is in your notes. A round-trip distance of 3000km and a speed of 300,000km seems to be inferring some simple numbers, neglecting that light travels slower in a non-vacuum.
With regards to part B, I think you are looking too deeply, it wants you to work out the overhead and do a simple calculation. Maybe talk to your tutor for clarification?
I ask because there are all sorts of issues here. For instance, if they are offering a 10Gbit service then it doesn't matter what the distance or medium is, the rate is 10Gbit. As you say though, TCP/IP throughput depends on datagram size/overhead, packet size/overhead, error detection / correction, line noise, permittable latency, the amount of data being moved (and so on and so on...) many many variables.
If it's undergrad computing, my guess is that most of the answer is in your notes. A round-trip distance of 3000km and a speed of 300,000km seems to be inferring some simple numbers, neglecting that light travels slower in a non-vacuum.
With regards to part B, I think you are looking too deeply, it wants you to work out the overhead and do a simple calculation. Maybe talk to your tutor for clarification?
Ok this is part of my mates answer:
10 GBps
3000km distance
300,000km/s fibre optic
125MB file
-------------------------------------------
TRANSFER RATE
Calculate sending the file:
propogation delay - transverse the medium:
distance/speed
3000/300,000 = 0.01 s
insertion delay - put into the medium:
max tcp window = 64kb x 1024 * 8 = 524288 bits
window/transfer rate
524,288/10,000,000,000 = .0000524288 s
Calculate receiving the acknowledgement:
propogation delay would be the same
0.01 s
insertion delay for acknowledgement(32bits):
32/10,000,000,000 = 0.0000000032 s
Add the times together:
= 0.020052432
Max transfer rate = window size / time
524,288/0.020052432 = 26145856 bps (26.2 mbps)
-------------------------------------------
FILE TRANSFER
convert 125MB into bits = 125 * 1024 * 1024 * 8 = 1048576000
file size in bits/transfer rate in bps
1048576000/26145856 = 40.1 s
On the right lines?
10 GBps
3000km distance
300,000km/s fibre optic
125MB file
-------------------------------------------
TRANSFER RATE
Calculate sending the file:
propogation delay - transverse the medium:
distance/speed
3000/300,000 = 0.01 s
insertion delay - put into the medium:
max tcp window = 64kb x 1024 * 8 = 524288 bits
window/transfer rate
524,288/10,000,000,000 = .0000524288 s
Calculate receiving the acknowledgement:
propogation delay would be the same
0.01 s
insertion delay for acknowledgement(32bits):
32/10,000,000,000 = 0.0000000032 s
Add the times together:
= 0.020052432
Max transfer rate = window size / time
524,288/0.020052432 = 26145856 bps (26.2 mbps)
-------------------------------------------
FILE TRANSFER
convert 125MB into bits = 125 * 1024 * 1024 * 8 = 1048576000
file size in bits/transfer rate in bps
1048576000/26145856 = 40.1 s
On the right lines?
For a) as well as the issues touched on i.e. window length, what about framing and tcp/ip packet overhead - thats effectively subtracted from your data rate yes? Also, I'm not sure whats wrong if anything but the answer of 26.2mbps sounds very low... allthoug 3000km fibre links aren't exactly my bread & butter so standing by for better explanation 
Even if there's not enough detail in the question to go into error rates, correction etc. as mentioned earlier, at least mention them in your answer - might be worth a few random ticks.
Even if there's not enough detail in the question to go into error rates, correction etc. as mentioned earlier, at least mention them in your answer - might be worth a few random ticks.
Agree with Zad - basically whoever wrote the question probably has never done this in the real world either 
theres about another dozen or so variables that are missing...one of the important ones (in real life) is the application thats doing the transferring - e.g. copying/pasting between Windoze platforms is more much more inefficent than say FTP etc - if you get something chatty with a lot of application turns its going to be pants anyway...
I think you need to read up in your notes and what/how you have been taught to calculate this stuff ...
...but I also think your mates answer is well out - Id be highly disappointed in real life if I couldnt get better than 26Mbps to 2 clients out of a 10Gbps link !!
..theres no where near enough information in the questions to do a real analysis so its got to be a basic maths calculation based on what/how youve been taught - without knowing that its hard to comment im afraid
in the real world to do this reliably, youd get replicas of the real servers and clients in the lab, capture traces of real user transactions over 10Gig-E then either use a modelling tool with the real measured params of the WAN link (OPnet ITGuru, Compuware Vantage etc) to predict what youre performance is going to be ...or just plain do it for real if you already have the link
theres about another dozen or so variables that are missing...one of the important ones (in real life) is the application thats doing the transferring - e.g. copying/pasting between Windoze platforms is more much more inefficent than say FTP etc - if you get something chatty with a lot of application turns its going to be pants anyway...
I think you need to read up in your notes and what/how you have been taught to calculate this stuff ...
...but I also think your mates answer is well out - Id be highly disappointed in real life if I couldnt get better than 26Mbps to 2 clients out of a 10Gbps link !!
..theres no where near enough information in the questions to do a real analysis so its got to be a basic maths calculation based on what/how youve been taught - without knowing that its hard to comment im afraid
in the real world to do this reliably, youd get replicas of the real servers and clients in the lab, capture traces of real user transactions over 10Gig-E then either use a modelling tool with the real measured params of the WAN link (OPnet ITGuru, Compuware Vantage etc) to predict what youre performance is going to be ...or just plain do it for real if you already have the link
You are probably right, in exam questions they generally look for results that are neat more-or-less round numbers.
The fact that they are using (propagation speed = speed of light) infers to me that they aren't looking at all the details. In the real world it is 0.6 to 0.8c in fibre. A 1500km fibre would also have repeaters every 10km or so which introduce their own delay. The time taken to generate data validation checks would dwarf the Ack insertion.
A long distance system wouldn't just send one 64kB packet at a time. It would stream a whole load of them out and then listen on the return channel for any acks and error reports.
If there are 2 separate tcp/ip units pumping data to the fibre, it may be that downloading 2 files takes no longer than 1 file? Thats the sort of nasty trick some lecturers can be looking for. I guess that's why it is a revision question. I would be interested in knowing what the official answer is
The fact that they are using (propagation speed = speed of light) infers to me that they aren't looking at all the details. In the real world it is 0.6 to 0.8c in fibre. A 1500km fibre would also have repeaters every 10km or so which introduce their own delay. The time taken to generate data validation checks would dwarf the Ack insertion.
A long distance system wouldn't just send one 64kB packet at a time. It would stream a whole load of them out and then listen on the return channel for any acks and error reports.
If there are 2 separate tcp/ip units pumping data to the fibre, it may be that downloading 2 files takes no longer than 1 file? Thats the sort of nasty trick some lecturers can be looking for. I guess that's why it is a revision question. I would be interested in knowing what the official answer is
tinman0 said:
The answer (surely) is 10Gbps. If you have contracted for a 10Gbps service then your max transfer rate should be pretty darn close to 10Gbps, otherwise the supplier is breach of contract.
In real life, yes. This, however, is the paralel universe call Academia. Another example being that i want a 5% payrise this year, but inhbitants of Academia want 25%.
I'd be interested to see what marks you'd get for giving that answer though. I dare you to put that on the exam paper!
Presumably they want you to calculate the minimum possible protocol overhead to derive the maximum possible throughput at the transport layer. I agree the speed-of-light stuff sounds like a red herring unless you're going to go into this in far more detail, which would require more information than they have given you. Do you know/have they told you anything about the MAC protocol on the fibre link? I assume you know how to work out the IP and TCP overheads once you know the MTU.
And even when you've worked it out, the Pointy Haired Boss won't pay for the bandwidth....
(I was asked a few weeks ago if it was OK to copy a 73Gb database from Chicago to London every night for backups. "Sure, so long as you don't mind started tonights before last nights has finished ..."
(I was asked a few weeks ago if it was OK to copy a 73Gb database from Chicago to London every night for backups. "Sure, so long as you don't mind started tonights before last nights has finished ..."
_dobbo_ said:
All very interesting, but does anyone who actually makes a living in networking need to know this sort of stuff? Then again, it shouldn't come as surprise to find that academic qualifications don't really prepare you for the real world!
LOL ...I used to make a very good living for a few years for a blue chip consultancy purely doing capacity planning, modelling and application profiling - £1.5k a day consultancy fees (for them...not me!) if you can predict whats going to happen before it does to a new network/application ...and get it right every time .......
edited to add: I have no academic IT related qualifications...they didnt do any back then
>> Edited by bogie on Monday 15th May 00:31
_dobbo_ said:
All very interesting, but does anyone who actually makes a living in networking need to know this sort of stuff? Then again, it shouldn't come as surprise to find that academic qualifications don't really prepare you for the real world!
Very very few people need to know this sort of stuff. I imagine the R&D people at hte comms providers would need it, and maybe their engineers. Designers and general networking engineers will just spec whatever line is required and leave the maths to BT or whoever.
zumbruk said:
The propagation delays are completely irrelevant (probably thrown in as red herring.) They're asking for transfer rates, not latentcy.
0/10 go sit in the corner with the dunce hat on
Propagation delays are VERY important in long distance (i.e. not in the same building) networking. Latency is the biggest problem for WAN engineers these days, not bandwidth. _dobbo_ said:
All very interesting, but does anyone who actually makes a living in networking need to know this sort of stuff? Then again, it shouldn't come as surprise to find that academic qualifications don't really prepare you for the real world!
All the good engineers keep this in mind. With the exam question the numbers don't sound relevant (they are) but when you start with the fact that London to Paris has an 8ms round trip time for packets.....
Ok had an email from my Lecturer:
Now who can explain CIDR to me in the easiest possible way! lol
Thanks for your help guys!
Reply said:
Q1 needs both propagation and insertion, though insertion will be very small compared to propagation and could be ignored without much loss of accuracy.
Part 2 – yes both in the same time. Other traffic not really relevant, but not a bad idea to cover that issue.
Now who can explain CIDR to me in the easiest possible way! lol
Thanks for your help guys!
agent006 said:
All very interesting, but does anyone who actually makes a living in networking need to know this sort of stuff? Then again, it shouldn't come as surprise to find that academic qualifications don't really prepare you for the real world!
Very very few people need to know this sort of stuff. I imagine the R&D people at hte comms providers would need it, and maybe their engineers. Designers and general networking engineers will just spec whatever line is required and leave the maths to BT or whoever.
I've always wondered who actually uses half the stuff taught in these things Question A said:
A company has offices some 3000km apart, and has taken up a Telephone Company offering a 10Gbps service. Calculate, with explanation, the maximum transfer rates for data links using TCP/IP over this service. You may assume that
a) the link is fibre-optic
b) the length of fibre between offices is exactly 1500km
c) The speed of light in fibre is 300,000km per second.
The question doesn't make sense, does it?
If the offices are 3000km apart, why is the length of fibre 1500km?
deva link said:
A company has offices some 3000km apart, and has taken up a Telephone Company offering a 10Gbps service. Calculate, with explanation, the maximum transfer rates for data links using TCP/IP over this service. You may assume that
a) the link is fibre-optic
b) the length of fibre between offices is exactly 1500km
c) The speed of light in fibre is 300,000km per second.
The question doesn't make sense, does it?
If the offices are 3000km apart, why is the length of fibre 1500km?
The round trip is 3000km.
Gassing Station | Computers, Gadgets & Stuff | Top of Page | What's New | My Stuff