Re: [dm-crypt] Re : Re : Poor performances with nfs and Kernel 3.x

[Milan Broz <mbroz@redhat.com>]

On 02/09/2012 08:37 AM, Arno Wagner wrote:
> On Wed, Feb 08, 2012 at 05:55:21PM +0000, Mickael wrote:
>> Here are more tests focusing the CPU load with htop snapshot:
>>
>> with Kernel 3.x, nfsd threads are using more CPU !?
>
> This is what I suspected. You setup is already slightly
> CPU-limited when reading/writing encrypted partitions.
> Any additional CPU needed by NFS is then directly deduced
> from CPU available for encryption.

Not sure if it help, but it would be interesting if you can try
to use patch from this thread
http://article.gmane.org/gmane.linux.kernel/1245748

(it can be completely unrelated though, I have no time to test
it now myself)

BTW I will add possibility to use null cipher to cryptsetup,
just to test exactly these situations (null cipher eliminates all
crypto operations), for now, you can try it with dmsetup and
measure preformance (this should uncover if encryption is
the problem or there is some timing problem):

(Replace DEV with your disk, note key is "-" - means empty)

dmsetup create x --table "0 $(blockdev --getsz DEV) crypt cipher_null-ecb-null - 0 DEV 0"

Milan