Hi Alan,
I have some small questions...
Have you ever tried a motherboard with i820 chipset and two NIC's?
I have a Asus P3C-D here (dual pIII 800 with 256MB rimm) and a D-Link
DFE570-TX NIC (quad DECchip 21143 behind a DECchip 21152 pci bridge).
The problem I'm seeing is extremely crappy pci performance.
With one of the NIC's active I see no real problems but when two NIC's are
active at the same time all hell breaks loose :(
if I generate traffic out via 2 NIC's at the same time I'd expect to get
about 2 x 100Mbit/s but in reality I get 2 x 25Mbit/s with this
motherboard. and with 3 NIC's active I get about 3 x 15-20Mbit/s.
I tried replacing the motherboard with an old SMP board based on the 440bx
chipset (cpus underclocked to 600MHz) and then I can easily get 3 x
100Mbit/s with 25-30% cpu used in the machine.
So my question is, have you ever seen such pci issues? I've tried all
BIOS settings I could find and a lot of diffrent kernels. I've also tried
the vanilla tulip-driver and the NAPI'fied one (which I have been helping
to test) and both show the exact same performanceproblems.
I tried routing a lot of packets and it started dropping a lot of packets
when the cpu was only 75% used according to both vmstat and cyclesoak
(with an UP kernel). A profile shows that default_idle gets about 25% of
the time. I assume the kernel is waiting for the pci bus.
And the last question, if you have the hardware, would you mind testing
something similar on the Asus A7M266-D? We've been thinking of getting a
few of these boards to replace the crappy i820 ones.
I saw that you included a patch for MPS 1.4 in -pre9-ac1, was this the
same patch you were talking about earlier and does it work?
/Martin
Never argue with an idiot. They drag you down to their level, then beat you with experience.
try to change slot
perhaps one of your nic is sharing irq with mouse usb ide
etc so change slot
check with more /proc/interrupts
On Tue, 12 Feb 2002, Martin Josefsson wrote:
> Hi Alan,
>
> I have some small questions...
>
> Have you ever tried a motherboard with i820 chipset and two NIC's?
> I have a Asus P3C-D here (dual pIII 800 with 256MB rimm) and a D-Link
> DFE570-TX NIC (quad DECchip 21143 behind a DECchip 21152 pci bridge).
>
> The problem I'm seeing is extremely crappy pci performance.
> With one of the NIC's active I see no real problems but when two NIC's are
> active at the same time all hell breaks loose :(
>
> if I generate traffic out via 2 NIC's at the same time I'd expect to get
> about 2 x 100Mbit/s but in reality I get 2 x 25Mbit/s with this
> motherboard. and with 3 NIC's active I get about 3 x 15-20Mbit/s.
>
> I tried replacing the motherboard with an old SMP board based on the 440bx
> chipset (cpus underclocked to 600MHz) and then I can easily get 3 x
> 100Mbit/s with 25-30% cpu used in the machine.
>
> So my question is, have you ever seen such pci issues? I've tried all
> BIOS settings I could find and a lot of diffrent kernels. I've also tried
> the vanilla tulip-driver and the NAPI'fied one (which I have been helping
> to test) and both show the exact same performanceproblems.
>
> I tried routing a lot of packets and it started dropping a lot of packets
> when the cpu was only 75% used according to both vmstat and cyclesoak
> (with an UP kernel). A profile shows that default_idle gets about 25% of
> the time. I assume the kernel is waiting for the pci bus.
>
> And the last question, if you have the hardware, would you mind testing
> something similar on the Asus A7M266-D? We've been thinking of getting a
> few of these boards to replace the crappy i820 ones.
>
> I saw that you included a patch for MPS 1.4 in -pre9-ac1, was this the
> same patch you were talking about earlier and does it work?
>
> /Martin
>
> Never argue with an idiot. They drag you down to their level, then beat you with experience.
>
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
>
On Tue, 12 Feb 2002, Davidovac Zoran wrote:
> try to change slot
> perhaps one of your nic is sharing irq with mouse usb ide
> etc so change slot
> check with more /proc/interrupts
Using an UP kernel with IO-APIC support and MPS 1.4 they don't share irqs:
CPU0
0: 66083638 IO-APIC-edge timer
1: 828 IO-APIC-edge keyboard
2: 0 XT-PIC cascade
4: 3 IO-APIC-edge serial
8: 0 IO-APIC-edge rtc
14: 17840 IO-APIC-edge ide0
16: 14 IO-APIC-level eth2
17: 12 IO-APIC-level eth3
18: 20790 IO-APIC-level eth0
19: 39 IO-APIC-level eth1
NMI: 66083576
LOC: 66081746
ERR: 0
MIS: 0
I've tried changing slots aswell and I havn't seen any improvments with
either MPS 1.1 or 1.4
/Martin
Never argue with an idiot. They drag you down to their level, then beat you with experience.
> Using an UP kernel with IO-APIC support and MPS 1.4 they don't share irqs:
try without IO-APIC support is it still the same ?
Z
>
> CPU0
> 0: 66083638 IO-APIC-edge timer
> 1: 828 IO-APIC-edge keyboard
> 2: 0 XT-PIC cascade
> 4: 3 IO-APIC-edge serial
> 8: 0 IO-APIC-edge rtc
> 14: 17840 IO-APIC-edge ide0
> 16: 14 IO-APIC-level eth2
> 17: 12 IO-APIC-level eth3
> 18: 20790 IO-APIC-level eth0
> 19: 39 IO-APIC-level eth1
> NMI: 66083576
> LOC: 66081746
> ERR: 0
> MIS: 0
>
> I've tried changing slots aswell and I havn't seen any improvments with
> either MPS 1.1 or 1.4
>
> /Martin
>
> Never argue with an idiot. They drag you down to their level, then beat you with experience.
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
>
On Tue, 12 Feb 2002, Davidovac Zoran wrote:
>
>
> > Using an UP kernel with IO-APIC support and MPS 1.4 they don't share irqs:
>
> try without IO-APIC support is it still the same ?
IIRC I've tried that too without any success :(
/Martin
Never argue with an idiot. They drag you down to their level, then beat you with experience.