[OpenAFS] low write/read performance with 1.8.x 1.9.0 client

Benjamin Kaduk kaduk@mit.edu
Tue, 13 Oct 2020 12:33:34 -0700


On Tue, Oct 13, 2020 at 09:26:43PM +0200, Giovanni Bracco wrote:
> Sorry for the typing mistake in  previous mail, but you got the point!
> 
> Now with the proper setting at the client startup the result is clear, 
> the low performance was indeed due to encryption on!
> 
> Thank you, the case is solved but it is indeed impressive that 
> encrypting the data transfer reduces performances of a factor 5 !

I don't think you should draw a generic conclusion about encrypting data in
general -- the fcrypt algorithm used by rxkad is spectacularly bad.  It is
based off DES, which in turn was specifically designed to be slow to
implement in software, and does not make use of modern processor features
like AESNI.  My testing during rxgk development was easily able to encrypt
traffic fast enough to saturate a gigabit link, on commodity hardware, with
an AES-based cipher.

-Ben