[Mnet-devel] patent-free LDPC code

[Bryce Wilcox-O'Hearn]

Jim Plank:

I'm excited to learn about this:

http://lists.mnetproject.org/pipermail/mnet-devel/2004-August/ 
003488.html

alternate URL:

http://article.gmane.org/gmane.network.mnet.devel/59

By the way, I am re-reading your excellent paper on practical use of  
LDPCs.  You used as your "slow" network a sustained rate of around 256  
KB/s download!  I am hoping to support people on crappy dial-up (lets  
say 4 KB/s) as well as to keep the door open to other weird transports.

I can infer from your results what would happen with "REALLY slow  
network" -- LDPC coding would do even worse when n is not >> 150, but  
RS coding would continue to go off-the-charts when n > 84.  (We use  
rate 1/3.)  This leaves us in a pickle, as what we most want is to set  
n = filesize / 2^16.  So for example a file of size 2^30 would yield n  
= 2^14.

Perhaps we should use different erasure codes depending on the file  
size.

It's too bad that there isn't an open source Cauchy implementation I  
could try.

Do you have any advice?

Thanks for your help.

Regards,

Zooko