[mnet-devel] fec news

[Zooko O'Whielacronx]

This guy named Enochian got on the IRC channel.

1.  Enochian says the badluckblockshares bug is obvious, and he has an 
explanation consistent with what we've observed.  Presumably this obviousness 
means that the fix is obvious, but I don't understand the bug.

2.  Enochian says that blocksharemod.c is probably as fast as can be -- in 
particular he thinks it is probably about as fast as my optimized fec.c will be 
(if I ever finish it).

3.  Enochian says that blocksharemod.c can easily handle m > 256.  There's an 
open question as to what the real-world performance cost is with e.g. k = 1024, 
m = 4096.

The reason that we care about k = 1024 is because then the file (or chunk) size 
could be 1 GB while the individual blocksizes that have to be transferred would 
be 1 MB.  

As far as I know, any erasure code has asymptotic runtime (for encoding) of 
l * k * (m - k), where l is the length of the file, m is the total number of 
shares generated, and k is the number of shares necessary (and sufficient) to 
decode.

Personally, I have very little time to spend on Mnet.  Maybe not even next 
Sunday (which is traditionally Mnet Hack Day).  If I do get time, I'll probably 
continue to fixate on fixing my optimized fec.c.  (Do NOT look at branch 
opt_fec.  It is a sadly non-working codebase that reveals a major error in my 
understanding of how the vandermonde erasure code works.  However, I now have 
fixed that error in my understanding.  I think.)

Regards,

Zooko



-------------------------------------------------------
SF.Net is sponsored by: Speed Start Your Linux Apps Now.
Build and deploy apps & Web services for Linux with
a free DVD software kit from IBM. Click Now!
http://ads.osdn.com/?ad_id=1356&alloc_id=3438&op=click
_______________________________________________
mnet-devel mailing list
mnet-devel at lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mnet-devel