cpuinfo, bogomips and duo core
Gordan Bobic
gordan at bobich.net
Tue Feb 12 13:37:27 UTC 2013
On 12/02/2013 13:24, Reindl Harald wrote:
>
>
> Am 12.02.2013 13:38, schrieb Gordan Bobic:
>>> -O3 -march=corei7 -mtune=corei7 -mmmx -msse2 -msse3 -msse4.1 -msse4.2 -maes -fopenmp -mfpmath=sse
>>
>> That just tells me you didn't push the machine to full saturation.
>>
>> Virtualization takes resources, and you cannot go faster by adding overheads. The only exception to this is where
>> you have hardware specifically designed for this, with a hypervisor in firmware running on proprietary resources
>> that a generic bare metal OS wouldn't be able to access anyway (or at least it wouldn't know what to do with it).
>> In such cases you have an added bonus that you can reboot the host without switching off the guests (and without
>> migrating them elsewhere, either). But hardware like that is very proprietary and uncommon.
>
> that just tells that you can disable a lot of services
> and overhead in a VM you would never do on bare metal
> and it tells that the hypervisor can schedule IO much
> more efficient as a generic kernel without less overhead
Utter nonsense. On bare metal, the Linux kernel scheduling even with
full awareness of the underlying CPU cores is pretty poor. C2Q is
particularly good example of this because latency and cache misses
between the two sets of 2 cores causes a 20%+ drop in throughput
compared to pinning heavy processes to a specific core. Systems with
multiple sockets also suffer from this issue particularly badly.
Now consider that you are effectively hiding the physical CPU layout
behind a hypervisor that applies it's smoke-and-mirrors and makes it
even harder for the guest kernel to do something reasonably sensible -
so you get another 20%+ overhead on top from the extra cache misses,
extra context switching overheads.
> however, it does not interest me to discuss with you
> after years of expierence with virtualization and bare
> metal for nearly any known type of servers with very
> dfifferent and mixed load
So much experience, so little understanding...
But you are right that it is off topic for this list. I shall not retort
further on this thread.
Gordan
More information about the users
mailing list