On Fri, Nov 1, 2013 at 8:12 AM, Prarit Bhargava <prarit(a)redhat.com> wrote:
On 10/31/2013 03:05 PM, Prarit Bhargava wrote:
>
>
> On 10/31/2013 03:02 PM, Josh Boyer wrote:
>> On Thu, Oct 31, 2013 at 2:33 PM, Bill Nottingham <notting(a)redhat.com>
wrote:
>>> Josh Boyer (jwboyer(a)fedoraproject.org) said:
>>>> (As for memory-critical cloud... I have no idea what that is to be
>>>> honest. All I hear from the cloud people is "smaller is
better".
>>>> Mostly that's image size, not memory overhead but I can imagine they
>>>> want that limited as well.)
>>>
>>> Admittedly, it's not the same as unswappable kernel memory, but I wonder
if
>>> for 2MB we can find that sort of working set size reductions in other places
>>> on the cloud image.
>>
>> Quite possibly so. I just hate to be wasteful if none of the 3
>> products clearly has a need. If 1024 is sufficient, we'll likely go
>> with that.
>>
>
> The reason I'm pushing 1024 as a target is that we had a previous request from
> users at SGI for a 1024. At least that is something we can point to instead of
> picking a value that no one really wants.
>
> IMO of course ;)
>
Memory usage data
difference between 1024 cpus and 128 cpus = 421k
OK. I'll change this in rawhide today.
difference between 4096 cpus and 128 cpus = 1.9M
I am now amazed that my memory actually had this right. That's rare.
Anyway, thanks much for getting these numbers. They really help.
josh