Re: [Xen-devel] [PATCH v2] Xen: Spread boot time page scrubbing across all available CPU's

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 30.09.13 at 17:42, Andrew Cooper <andrew.cooper3@xxxxxxxxxx> wrote:
> On 30/09/13 16:35, Jan Beulich wrote:
>>>>> On 30.09.13 at 15:56, Malcolm Crossley <malcolm.crossley@xxxxxxxxxx> 
>>>>> wrote:
>>> The difference between 1 core per node bandwidth and 3 core's per node 
>>> bandwidth is:
>>>
>>> AMD: 30000MB/s (1-core) vs 48000MB/s (3-core)
>>> Intel: 12000MB/s (1-core) vs 38000MB/s (3-core)
>>>
>>> So I think it's worth the extra complexity to have multiple core's per 
>>> node scrubbing memory.
>> The numbers are convincing. Which means that the only request
>> I'd have is to avoid more than one hyperthread on a core to get
>> picked for doing the scrubbing.
> 
> Why? Being independent operations, scrubbing like this will still be
> sped up by hyperthreading.
> 
> As scrubbing RAM is simply a race to get it done as quickly as possible,
> unless it can be demonstrated that using all cores is actively
> detrimental to the overall time taken, we really should use all cores
> where possible.

As long as the actual scrubbing routine is written well enough
(to saturate the store bandwidth of a single core), having
two threads compete with one another for the store ports can
only be negatively affecting performance. _If_ multiple hyper-
threads would really turn out to be beneficial, that would thus
rather be an indication to improve clear_page() (or whatever is
being used for the actual scrubbing).

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel