'Serious' crypto?

[Matthew Seaman]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 27/05/2010 21:49:12, Peter Cornelius wrote:

>> NAT.  Doing serious crypto slows things up somewhat.
> 
> I've been pondering this since a while but thought that crypto 
> engines on modern hardware would make 'extra' hardware accelerators
> obsolete?

Yes -- in many use cases this is true.  Modern processors are fast
enough that they don't need an external accelerator to perform.  It
doesn't mean that running crypto imposes *no* extra cost on a server.
For instance, a web server running HTTP will (roughly speaking) be able
to support an order of magnitude more simultaneous sessions than the
same site served over HTTPS.

> Or is it still worthwhile to consider hardware accelerators such as 
> the ones guys like soekris [1] and others offer? Does anyone have an 
> idea "how much" such an accelerator may help on older vs. on newer
> hardware?

Those soekris boards are designed to work in low power (both in wattage
and in compute capability) appliances.  That is a perfectly viable
alternative design for a crypto-gateway router / packet filter intended
for traffic levels within the specification they claim.

Hmmm... 250Mb/s IPSec throughput is (I think -- not having tried this, I
cannot be certain) easily accessible through a fairly run of the mill
server such as the HP Proliant DL120 G6.  Of course, the HP box costs
about 4--5 times as much as the Soekris. It will have a great deal more
spare RAM, disk, compute capacity etc.  No idea abut on-going support
costs, but I don't think you could get support cover with a 4 hour
on-site response from Soekris...

> Would multiple engines work (and help) at all? From crypto(4), I 
> would not guess so. One consequence would be that there may be
> certain limitations in using a separate accelerator once the platform
> comes with its own accelerator device?

One feature that hardware accelerator boards provide which is hard to
get otherwise is plenty of random numbers on tap.  Generating
cryptographically strong randomness in volume is pretty hard
computationally, and a hardware solution really helps things like IPSec
throughput.

Also, if you need really high volume crypto traffic throughput (multiple
Gb/s levels), then yes, you will need specialised hardware.  However, in
this case, you're likely to be using pretty fancy routers (Cisco,
Juniper, etc.) and those all have options for hardware acceleration
built into interface cards.

	Cheers,

	Matthew

- -- 
Dr Matthew J Seaman MA, D.Phil.                   7 Priory Courtyard
                                                  Flat 3
PGP: http://www.infracaninophile.co.uk/pgpkey     Ramsgate
JID: matthew at infracaninophile.co.uk               Kent, CT11 9PW
-----BEGIN PGP SIGNATURE-----
Version: GnuPG/MacGPG2 v2.0.14 (Darwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAkv/c3QACgkQ8Mjk52CukIxJIwCbBTN1wcUcOodn6s7Sxa8yv4lE
d+sAmwTZLxLo7KyMIdEKJJOLfa8OfVmI
=KzX7
-----END PGP SIGNATURE-----