Overview

Midterm Review: Q1

My network, which is very slow, has a latency of 50 s and a throughput of 1 Kb/s (10^3 b/s). If I am sending 10 Kb (10^4 bits) of data at maximum speed, and if I start to send when my clock shows 12:30:00, what time will my clock show when the receiver gets the last bit of data? Show your work to get credit.

A: 10 s for 10Kb (at 1Kb/s), plus 50s = 60s. Clock time 12:31:00.

Q2

I have received a packet with bit errors given by the pattern 0001000001000010000110000. Will my CRC be sufficient to detect the error? Explain why or why not, or what additional information you would need in order to be able to state whether or not the CRC would detect the error.

The error pattern has a 1 for each bit that is in error (changed from the original message), and a 0 everywhere else. Therefore, this error pattern has an odd number (5) of bit errors, which a CRC will detect.

Q3

I am starting a company to produce a new kind of networking product, the switch-hub (the shwub). This product is more efficient than a hub because (like a switch) it can pass messages between different ports simultaneously (so, for example, an ethernet packet can be sent from port 2 to port 5 at the same time as another packet is sent from port 3 to port 1). A shwub is less expensive than a switch because I am leaving out the forwarding table, the table in the switch that stores (ethernet address, port) pairs. Will I find lots of eager customers? explain why or why not.

Because the shwub is missing the forwarding table, it must broadcast each message on all its ports (the switch uses the forwarding table to decide where to send each packet). This makes it no more useful than a hub, and I am not likely to have a lot of customers.

Q4

If voice samples are received at the rate of one byte every 125 us (i.e. 8,000 bytes per second), how long does it take to accumulate enough bytes to fill

Q5 -- insight

This problem explores two ways of extending the ethernet protocol to work over long distances.

Suppose I change the protocol so all packets are the same length (20,000 bits), I split the channel into 20,000 subchannels, and send all the bits in parallel, each over a different channel.

If I split the channels into subchannels, my total capacity is likely to remain unchanged. That is, it will take me as long to send one bit on each channel as it would take me to send the 20,000 bits on a regular ethernet.

Q5/a

Does this modification overcome the length limitations of Ethernet? Why or why not?

The length limitation of ethernet is due to the "length" of a minimum-sized packet -- to provide collision detection, the packet must reach the far end of the network, and the jamming signal must return to the sender, before the packet has been completely transmitted. By splitting into 20,000 channels, the "length" of a packet is the same.

Q5/b

Now modify this protocol so instead of sending the bits in parallel, I pick one channel at random and send the bits sequentially on this channel. Does this work over long distances? Why or why not?

This actually helps, because the length of the packet is much longer (since each bit is transmitted more slowly).

Q6

My ATM switch has a hardware problem that causes it to drop 1 bit every 1,000,000 (that is, the 1,000,000-th bit is never sent) -- the remaining 999,999 bits are sent correctly.

I am receiving 10,000-bit AAL-5 packets.

  1. what kind of errors does my ATM report? places where ATM detects these errors include: the header CRC of the cell following this one (framing error), the AAL-5 CRC, the AAL-5 length field.
  2. the bit error rate is 10^{-6}. What is the packet error rate? Since one packet in 100 is corrupted, the packet error rate is 10^{-2}.
  3. if I switch to using 100,000-bit (12,500-byte) AAL-5 packets, the packet error rate is 10^{-1}.

Q7

Internet History

Internet Survivability

Internet Architecture: Protocols

(see also picture)

Internet Architecture: Networks