[Vortex] Is vortex assuming that msg numbers on channels start at 1 and increase by 1/

Francis Brosnan Blazquez francis at aspl.es
Thu Nov 27 13:16:53 CET 2008


Hi Sam,

> Beep performance can be much poorer. Its essential to use window sizes
> in the close too 100K range for bulk transfer, and to make sure you
> fill the window with data, probably pre-sending 5-10 messages of about
> 10K each, to come close to TCP. And disable NAGLE :-)

Just to confirm Vortex disables Nagle's algorithm by default [1].

> I'm sure BEEP can do much better, there is no reason in the protocol
> to be slow, its all in how you use it. I hope! 

I can confirm it is. We have tested raw TCP, ftp, http and scftp
transfers rate compared with a Vortex Transfer and we are performing
nearly to the perfect efficiency (really close to raw TCP). 

Obviously you have to reduce to the minimum protocol headers by
increasing frame size (VortexChannelFrameSize) and update receiving
channel window size (vortex_channel_set_window_size).

Another users have also confirmed high transfer rates on highly
responsive connections [2]. So, there is no base to think you'll have a
considerable penalty if you use BEEP instead of raw TCP. 

> I'll write up
> recomendations when I have them. Still doing testing.

Good. I think there are lot of work in this area to detect a propose a
proper balance between all factors that control a BEEP transfer. 

Cheers!

[1] http://www.aspl.es/vortex/btf.html
[2] http://lists.aspl.es/pipermail/vortex/2008-October/000664.html

> Cheers,
> Sam
-- 
Francis Brosnan Blazquez <francis at aspl.es>
Advanced Software Production Line, S.L.



More information about the Vortex mailing list