TLB shootdowns in /proc/interrupts (different results)

[Kevin Wilson]

Hi,
I have two x86_64 machines:
On the first there is Fedora 18 (x86_64); on the second there is
Ubunutu 12.10 (server, 64 bit).

I run cat /proc/interrupts on both machines; in both cases, I run this command
after an hour when the machine is up.
I get:
On the fedora machine
cat /proc/interrupts | grep TLB
TLB: 0 0 TLB shootdowns
On the Ubuntu machine, however, there are many TLB shootdowns, on both
cores.


Can anyone explains why is it so ?

rgs,
Kevin

[Mulyadi Santosa]

Hi...

Post by Kevin Wilson
Hi,
On the first there is Fedora 18 (x86_64); on the second there is
Ubunutu 12.10 (server, 64 bit).
I run cat /proc/interrupts on both machines; in both cases, I run this command
after an hour when the machine is up.
On the fedora machine
cat /proc/interrupts | grep TLB
TLB: 0 0 TLB shootdowns
On the Ubuntu machine, however, there are many TLB shootdowns, on both
cores.

That is a very interesting data....

btw, after quick check at
http://stackoverflow.com/questions/3748384/what-is-tlb-shootdown, I
conclude that likely Fedora 18 default kernel does something, that in
your workload prevent tlb shootdown. What can that be? IMHO, one high
possibility is preventing process rescheduling to another core(s)
and/or grouping all threads in same group in same core(s).

But I am not sure. That sounds like defeating the purpose of SMP.