Re: Udp sending performance in Gbit Ethernet

JTL wrote:
[..]
> I again kindly ask you to read the previous posts and check the previous
> threads from google. This phenomenon has been detected in multiple
> configurations and multiple windows versions.

Writing network drivers is my main job. I did that for various network
cards since 1988.

We always did a lot of testing and thinking wrt performance
optimization both for hardware and software. One thing you learn is
that the more complex your optimization approach gets, the less
efficient it usually also gets (paradoxal, I know). There are just too
many side effects one cannot think of.

So I am *not* actually suprised by the behaviour you describe. We often
saw a sudden drop of perfromance for certain packet (frame) sizes. And
there can be just so many reasons like interrupts and DMA.

If the number of interrupts increases by a factor of 3 as you describe,
then there is probably some limit reached in the driver or in the card
such as the maximum DMA block size or alike, which forces the driver to
split frames into several DMA transfers. Since some cards generate an
interrupt at the end of each DMA transfer, this would be an
explanation.

But this is just wild guessing. I am not familiar with the actual
architecture of neither the driver or card's hardware that you use.

All you can do is:

1. Try different versions of the driver for the card (even older ones).
2. Try some other card (different chipset) along with its driver.
3. Holler at the card vendor's support (don't expect too much).

BTW, which chipset does your GigE card use (Marvell, Broadcom, Intel,
etc.)?

HTH, Stephan

.

Relevant Pages

  • Fwd: Re: [RFC] pata_icside driver
    ... we need to disable the interrupts from the port ... on the card if no drives are found, ... This driver continues to have the so far ignored issue concerning ... * Purpose: disable interrupts from card ...
    (Linux-Kernel)
  • Re: Preferred method to get interrupts
    ... I had problems with interrupts ... On the X51 when the card is inserted, the card is powered up, but stays ... We got are driver to load in a similar fashion to yourself but what I found ...
    (microsoft.public.windowsce.platbuilder)
  • Re: Preferred method to get interrupts
    ... I think the X51 CF problem stems from Dell trying to implement the new PC ... CARD structure, instead of what HP did and remove it from the OS. ... We got are driver to load in a similar fashion to yourself but what I ... I have an IPAQ hx2495 but have been unable to get interrupts on that ...
    (microsoft.public.windowsce.platbuilder)
  • Re: Udp sending performance in Gbit Ethernet
    ... >> saw a sudden drop of perfromance for certain packet sizes. ... >> such as the maximum DMA block size or alike, which forces the driver to ... >> split frames into several DMA transfers. ... Try different versions of the driver for the card. ...
    (microsoft.public.development.device.drivers)
  • device driver race conditions problems (kernel 2.4)
    ... I'm working in a device driver for kernel 2.4 to use with our new CTI ... My driver must be able to manage more than one card using the same ISR ... interrupt cannot interrupts the other but when one reaches another, ... static inline void inc_pointer_signal(volatile unsigned int *index, ...
    (comp.os.linux.development.system)