ilovepi / ncd Goto Github PK

There is a sloppy busy wait in the recv4 () function. It should use a condition variable instead.

Fork the project, and move all the new framework and library code into master

Ansible will allow a great deal of the scripted logic to be replaced with Ansible code and playbooks.

Consider looking into the Geni Experiment Engine's use of Ansible for ideas

Output timestamps instead of elapsed round trip time

open and close a connection via the kernel, and use raw socket to simply send our packets... makes life easier, also will be harder to detect.

have the program determine if network compression or the presence of some other middlebox inspecting, or modifying packets is likely

Probably use the dime difference as a %

-maybe add another commandline option to set the sensitivity of the estimation

Either remove Raw UDP communications, and let the kernel determine the source IP, or find an appropriate way to determine the source IP without opening another socket

The testbed should have its own git repo, and should be a submodule within NCD

Add support for a fully connected tcp connection, ie full handshake.

Refactor ncd so it can use different ports and payloads for each packet train

add a function that can do the full tcp close properly

The python and pandas code has performed well, however, Julia and Gadfly gave superior results.

Consider patching the Julia Pcap package to handle endianness correctly, and formalizing the scripts

Need a man page and more/better docs
Add better comments to codebase
improve documentation

Ensure that all bounds on buffers are properly checked

The classes and functions should be organized in a more organic way to form a separate library from the actual clients and servers

Take a look at where Error codes are returned and checked, and ensure that the program handles them appropriately, and notifies the user what the problem was.

Find places where the code is slow

-profile the performance, and identify bottlenecks
-try to improve access patterns for better caching

• File for fill_data()
• Time between first and second test

Find any unused functions and remove them
mkudphdr() - maybe useless since we don't manually set up the udp header anymore

There's no checking for user defined files for fill_packet(), there needs to be some error checking, if the file exists, if its too short, too long, etc.