On Wed, Dec 06, 2006 at 04:51:51AM +0100, Max Laier wrote:
On Wednesday 06 December 2006 01:17, Luigi Rizzo wrote:
...
First, this proposal, with 36 multiplies and one division, the
function seems rather expensive for e.g. a low end cpu (arm or
soekris) as you might find on
On Wed, Dec 06, 2006 at 04:51:51AM +0100, Max Laier wrote:
I tried the reference machines (see hacked up attachment):
78x ia64
40x amd64
60x p3
16x p4
I don't have my Soekris set up, so if somebody could give it a try.
On my 4.11 Soekris 4501 box, the test shows about 70x for gcc -O2
and
On Wed, Dec 06, 2006 at 10:56:42AM +, David Malone wrote:
On Wed, Dec 06, 2006 at 04:51:51AM +0100, Max Laier wrote:
I tried the reference machines (see hacked up attachment):
78x ia64
40x amd64
60x p3
16x p4
I don't have my Soekris set up, so if somebody could give it a try.
On Wed, Dec 06, 2006 at 01:29:31AM -0800, Luigi Rizzo wrote:
the top forwarding performance of a soekris is around 30-35kpps if
i remember well - this translates in around 30us/packet all included.
Is that the peak with ipfw2, IPv6 packets and dynamic rules turned on?
as you see from the
On Wed, Dec 06, 2006 at 11:38:47AM +, David Malone wrote:
On Wed, Dec 06, 2006 at 01:29:31AM -0800, Luigi Rizzo wrote:
the top forwarding performance of a soekris is around 30-35kpps if
i remember well - this translates in around 30us/packet all included.
Is that the peak with ipfw2,
maybe just ipfw2-ipv4 on and a single accept rule.
that's to give an estimate of all the remaining packet processing
costs that you were mentioning (interrupts etc.)
OK.
I've read the description of the Hsieh hash now and I'm pretty sure
it should be possible to produce lots of collisions
Hi,
with a lot of help from David Malone and JINMEI Tatuya we came up with the
following hash function for IPv6 connections using universal hashing.
Note that while it looks a lot more complicated, it is unlikely to
consume (much) more time. The most expensive operation is still the
memory