Re: freebsd router



--- "Matthew D. Fuller"
<fullermd@xxxxxxxxxxxxxxx> wrote:

> On Wed, Jan 11, 2006 at 07:22:15AM -0800 I
> heard the voice of
> Danial Thom, and lo! it spake thus:
> >
> > You're basically limited in an MP environment
> to sucking the packet
> > out of the ring and throwing it into a queue.
>
> I can only presume you're being willfully
> silly. An MP environment is
> quite capable of grabbing the packet from an
> interface, and then
> CONTINUING TO PROCESS IT while the other CPU
> grabs the next packet and
> starts processing it. You can process through
> to completion JUST as
> easily on a MP system as you can on a UP
> system, except that you're
> not blocking the next packet while you do it.
> Furthermore, you've got
> another CPU to do things like update the
> routing table and deal with
> OTHER interrupts and kernel tasks at the same
> time.

and another concept you fail to grasp is that
with 1 processor the data for the entire
transaction sits in the cpu cache, whereas with 2
it has to be read again. Its not necessarily
desireable to have 2 processors working on the
same data. There are a lot of issues that aren't
as black and white as "2 processors must be
faster than 1".

DT

__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com
_______________________________________________
freebsd-isp@xxxxxxxxxxx mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-isp
To unsubscribe, send any mail to "freebsd-isp-unsubscribe@xxxxxxxxxxx"

Relevant Pages

  • Re: dummynet, em driver, device polling issues :-((
    ... 32 bit 33MHz PCI bus, the math just doesn't work for high speeds. ... packet was finished being received, ... that the CPU isn't the bottleneck there. ... >> Nothing fancy on polling, i think we set HZ to 10000 ...
    (freebsd-net)
  • Re: CPU utilisation cap?
    ... CPU utilisation always peaks at ... > program (by inserting additional packet copies), CPU utilisation does ... a sampling process that polls the kernel's notion of CPU% measurement ...
    (freebsd-performance)
  • Re: packet drop with intel gigabit / marwell gigabit
    ... packet drop with intel gigabit / marwell gigabit ... cause packet loss, it is difficult to get insight of the problem. ... when I said that "cache design affects memory bandwidth " is very vague. ... Generally, if we talk access data between CPU and main memory, then technically is right. ...
    (freebsd-performance)
  • Re: TCP/IP send, sendfile, RAW
    ... the platform is a netsilicon NS9360 witch include an 100MBit ethernet device. ... CPU load is 100% when running sender program. ... What you have to consider is processing required per packet. ... At the interrupt level you could use NAPI or interrupt coalescing etc. ...
    (Linux-Kernel)
  • Re: intel 64 bit instructions vs amd 64 bit instructions
    ... HyperTransport doesn't need more pins to transport wider addresses. ... packet header. ... However the current AMD CPUs don't support that. ... maximum amount of RAM you can connect to a CPU. ...
    (alt.lang.asm)