Anton,
Looks like you don't have as much interrupts to trigger the bug on pre-MCLGETI.
Take a look at PR 6419 and our box also can't survive 15k with re(4)
and rock stable on 70k with bnx(4).
OpenBSD 4.7-beta (GENERIC.MP) #82: Fri Feb 5 01:05:44 MST 2010
[email protected]:/usr/src/sys/arch/amd64/compile/GENERIC.MP
re0 at pci4 dev 0 function 0 "Realtek 8168" rev 0x01: RTL8168 2
(0x3800), apic 2 int 17 (irq 11), address 90:e6:ba:b6:ec:6e
Alexey
Anton Maksimenkov wrote:
> >Description:
> When system reaches thousand connections, machine becomes unresponsive.
> Though it's possible to break into ddb (stack traces differ from time
> to time).
>
> Original report: http://marc.info/?l=openbsd-misc&m=128426958003630&w=2
> >How-To-Repeat:
> Run any network stress program that about a thousand connections.
> >Fix:
> Use pre-MCLGETI version of the driver.
