[OpenAFS] Why is speed of AFS loopback adapter set to 10 Mb, even if physical interface is Gb capable?

[Lars Schimmer]

On 25.04.2010 16:19, Holger Rauch wrote:
> Hi Lars,
>=20
> thanks for your reply.
>=20
> Lars Schimmer schrieb am Sunday, den 25. April 2010:
>=20
>> Usual auto negotiation does its job quite good.
>> But OpenAFS enables encryption of data transfer on each standard insta=
ll
>> - which results in drastic performance decreases.
>=20
> Thanks for mentioning this and that's exactly what I'm wondering
> about. Encryption too is involved when doing SCP transfers (even
> stronger encryption, such as Blowfish or AES 256) and I get scp speeds
> of 60 to 82 MB/sec when transferring to native ext3 fs.
>=20
> So, my question is: Why is there almost no decrease when using SCP
> with a strong encryption algorithm (compared to plain DES that AFS
> uses) but there is *drastic* decrease in performance with only plain
> DES enabled compared to an unencrypted transfer?

OpenAFS does not use DES. And those are complete different
implementations, as already laid out by simon.
Data is far often copied and shuffled around within OpenAFS than with
scp/ftp/ssh.
And the encryption implementation was not designed to be done perfect
over gigabit - it was designed for 1-2 mbit lines.
A new implementation of that encryption is currently worked on, guess it
will be faster.
Oh, btw, a lot of small files will still fail to reach higher speeds, as
callbacks are done massive. A network filesystem has some more needs
than a local one ;-)

>> E.g. my workstation client (linux) does 10-12 MB/sec with encryption
>> enabled, but >70 MB/sec with encryption disabled.
>> My point is on: disable encryption.
>> Btw, our linux clients does roughly 20-50 MB/sec so far. Still need to
>> test memcache.
>=20
> Any good ideas as to how to test memcache?

Get a machine use memcache, deactivate encryption and test the transfer
rates?
Or get a machine with those nice SSD and set the cache with ~10-30 GB on
those fast SSD. Should increase bandwidth, to.
I just need to set that client up.
Although current rates are quite fine, still got most machines on
100MBit switches and those lines are saturated with OpenAFS.

> Thanks in advance & kind regards,
>=20
>=20
>    Holger
>   =20


MfG,
Lars Schimmer
--=20
-------------------------------------------------------------
TU Graz, Institut f=FCr ComputerGraphik & WissensVisualisierung
Tel: +43 316 873-5405       E-Mail: l.schimmer@cgv.tugraz.at
Fax: +43 316 873-5402       PGP-Key-ID: 0x4A9B1723