[Mnet-devel] erasure code optimization project

[Lenny G Arbage]

--- Bryce Wilcox-O'Hearn <zooko at zooko.com> wrote:

> What happened when you ran "make" in the rizzo
> subdirectory?

  I just figured it out -- you should remove 'fec.S'
from the repository so that 'make' rebuilds it for the
appropriate host architecture.  As it was, I got a
slew of messages from the assembler about "no such
instruction" etc.

  Here are the numbers I get for the same file as
reported before, using rizzo:

time ~/tmp/ecopt/rizzoerasureimpl/rs_encode_file r1 84
170 r1.rizzo.coded/x > /tmp/rizzo.decoder
 
real    0m0.147s
user    0m0.090s
sys     0m0.010s

time ~/tmp/ecopt/rizzoerasureimpl/rs_decode_file x >
r1.decoder 2>decode.log
 
real    0m0.056s
user    0m0.030s
sys     0m0.010s


  Rizzo is certainly faster than plank on my machine,
but LDGM still beats it by more than a factor of 5x...
 Of course, this is just one datapoint.  I'll run on a
few more filetypes/sizes to see how things hold up.

  -Lenny


		
__________________________________
Do you Yahoo!?
Yahoo! Mail is new and improved - Check it out!
http://promotions.yahoo.com/new_mail