Reliable data transport implemented over UDP.
- Quan Do, biquando@g.ucla.edu, 105490012
- Theodore Lau, teddy1405@g.ucla.edu, 405462550
We modified the while(1) loop in server_src/server.cpp to additionally send
appropriately numbered ACKs and dup-ACKs and log them in its window. We
modified client_src/client.cpp to
- keep filling its
windowwith data packets until thewindowis either full or we reach the end of the file, - send the data packets it creates,
- tell its
windowwhich packets have been ACKed, - and tell its
windowto resend any packets which have timed out.
All other common functionality was factored out into common_src/, including
some repeated template code and our Window class.
Our Window class implements the bulk of our packet management, including
- tracking packet statuses (waiting, ACKed, inaccessible),
- tracking packet timeouts,
- shifting the window,
- and manipulating the circular buffer.
The same Window class is used in both server_src/server.cpp and
client_src/client.cpp to realize Selective Repeat.
Our first "major" problem was not realizing the template code polled the sockets instead of using blocking, but after noticing it, we were quickly able to adjust our logic appropriately. Our second major problem was synchronizing the sequence and ACK numbers between the server and client, especially for the teardown stage and for handshakes in which packets were lost. We realized we could solve the problem by simply updating one host's variables based on the latest packets the other host has sent.
Our third problem involved the contents of our data packets, specifically
concerning the very last data packet. We use feof() in client_src/client.cpp
to check whether we have already read the entire file we are sending, but it
only returns non-zero after a failed read. Therefore, we would create and send a
packet of payload size 0 after already sending the entire file, which is
incorrect behavior. We fixed it by simply ignoring any size 0 reads.
Our last major problem involved making sure our printouts simply complied with what the spec and autograder expected, which involved tweaking each of our print calls slightly until our output matched what was expected.
We used no outside resources other than what was presented in class and discussion.