[OpenAFS-devel] rxkad performance

[Love]

So I've started to think about turning on crypt on all my clients, so I
check the how much the performce would go down if I turned it on. So I
checked it with my perfomance testing cache-manager (afsfsperf in current
arla, perf in older versions)

The numbers are here:

	/afs/su.se/home/it.su.se/l/h/lha/Public/su-bench/

cpu usage: data is the reading and writing part, meta is the
creation/removal of files.

The -bg version of afsfs8-1 is the rxkad from arla hack into openafs. As
you can see from the number in the directory above and from fc_test below
Björn's fcrypt is faster the OpenAFS version.

: root@hummel ; ./openafs-i386_fbsd46/src/rxkad/fc_test 
fc_keysched    = 0.20 us
fc_ecb_encrypt = 0.59 us
fc_cbc_encrypt = 3.73 us
: root@hummel ; ./openafs-i386_fbsd46-bgrxkad/src/rxkad/fc_test 
fc_keysched    = 0.17 us
fc_ecb_encrypt = 0.22 us
fc_cbc_encrypt = 1.37 us


I think you should consider switching rxkad module, or if not, just switch
the fcrypt part.

The fun part of switching to bg's rxkad is that its infested with krb_
stuff, so one need to rip those equivalent out of openafs rxkad.

The source where I ported bg's rxkad to openafs, but it need some cleaning
up before using it. It also needs to have the parts the I stole from
openafs rxkad marked as such (mostly in compat.c). I sent the et file into
bitbucket because I'm lazy.

        /afs/su.se/home/it.su.se/l/h/lha/Public/bg-rxkad.tar.gz

Love