Hi, Adrian, Graeme and Thomas
Thank you for really really rapid reply !
What an incredible ML !
> What network card are you using on the director? There's only so many
> interrupts you can generate from one NIC before the CPU wedges - and in
> most cases every packet generates in interrupt. If you're using
> plain-old-PCI, you get less throughput than more advanced PCI variants.
> There was a thread about this recently. Search the archives for it.
the device is following
eth0: (PCI Express:2.5GB/s:Width x1)
eth0: Intel(R) PRO/1000 Network Connection
>This is more in line with hardware / software limitations than with a
>bug in your config. The situation I was describing happened at very low CPS.
>As has transpired on the ML, you can try to boost your performances by:
>- upgrading the kernel to a 2.6.20
>- change your network cards to better models that can handle more than
>one packet per interrupt.
I see.... I would have liked to avoid the problem with kernel
parameter tuning if possible...
But, I'll try the above.