Re: Livelock limit engaged?

[Erik Wikström <erik-wikstrom@xxxxxxxxx>]

On 2007-08-11 01:24, Dave Hayes wrote:
> Matthew Dillon <dillon@apollo.backplane.com> writes:
> >  run vmstat -i
>
> # vmstat -i
> interrupt                   total       rate
> clk                                              281783        437
> atkbd0                                                4          0
> sio0                                                  0          0
> fdc0                                                  0          0
> ppc0                                                 18          0
> acpi0/em0                                             0          0
> uhci0                                                 0          0
> uhci1/uhci2/uhci3/twe0/fxp0/atapci1/atapci1     6766087      10506
> ata0                                              12463         19
> ata1                                                  0          0
> irq19                                               101          0
> swi_siopoll                                           0          0
> swi_crypto/swi_camnet                                 0          0
> swi_cambio                                            0          0
> swi_vm                                                0          0
> swi_taskq                                             0          0
> Total                                           7060456      10963
>
> Ok I see the interrupt in question. (I don't suppose there is any way to
> move the USB interrupts to another less used IRQ?)
>
> >    Monitor your interrupt rate with systat -vm 1
> >    If the interrupt rate is normal for the workload on the machine,
>
> There's a good question. Define "normal". :) I have no idea what normal
> is, this is a brand new box.
>
> Here's a vmstat -i from a production machine that is doing your basic
> web/email/DNS:
>
> # vmstat -i
> interrupt                   total       rate
> clk                                           784643353        281
> atkbd0                                                1          0
> sio0                                                  0          0
> ppc0                                                  1          0
> acpi0/em0                                             0          0
> uhci0                                                 0          0
> uhci1/uhci2/uhci3/twe0/fxp0/atapci1/atapci1    75223514         26
> ata0                                              71880          0
> ata1                                                  0          0
> irq19                                                96          0
> swi_siopoll                                           0          0
> swi_crypto/swi_camnet                                 0          0
> swi_cambio                                            0          0
> swi_vm                                                0          0
> swi_taskq                                             0          0
> Total                                         859938845        308
>
> The two machines are supposed to be identical, but obviously they
> aren't. (The last one was bought recently, the previous one in April.)
>
> If I do your systat -vm 1, I hardly see a peak interrupt rate on
> the production machine, but the new machine spikes to 30-40%. 
>
> Both motherboards are D946GZIS Intel, identical as far as I can 
> tell from the model and packaging. 

Just a guess (since this is not my area) but have you checked if they 
have the same BIOS revision? It might be possible that the newer one 
have an updated BIOS.

--
Erik Wikström