Interrupt Affinity in SMP

From: Hari LKML
Date: Wed Jun 30 2010 - 02:08:25 EST


Hi All,

We have HP DL380 Server with 48GB of DDR3/1333 MHz RAM, 4 1Gbps NIC
Ports, 1 External RAID Card, 1 Internal RAID Card and 2  Intel(R)
Xeon(R) CPU X5570  @ 2.93GHz used as an NFS storage server. Running
Ubuntu 9.10 2.6.31-14-server.
(The Storage is used as an NFS storage for more than 300VMs running
using Citrix XenServer)

For More info on the Hard ware configuration
http://h18000.www1.hp.com/products/quickspecs/12028_div/12028_div.html

With HTT enabled in BIOS i can see 16 CPU(0-15) as per cat /proc/cpuinfo.

The problem is all the interrupts are handled by CPU0, as per
/proc/interrupts including the timer interrupts.
And the value of /proc/irq default_smp_affinity is ffff.
Also for all the interrupts the smp_affinity is ffff.

What i believe is with these setting all the interrupts should be
handled by the First 8 CPUs in a Round robin Fashion.
Or in our case at worse by the first 4 cores (since HTT is enabled -
*not clear with this concept though some once please put some light
over here also for better understanding :-)* ).

Also i tried to edit the smp_affinity of an interrupt to a value 0004
to test that if the interrupt is handled by some other core. But i was
unable to edit this value in a running system i got the access denied
error so i can't do this.

Please help me in distributing the interrupts among the other cores of
my system which might increase some performance of my system.

let me know if you need some log of the system or some other info
about the system.

Hari
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/