Re: Dell R930 server

2016-11-09 Thread Theo de Raadt
>What about my questions ?

the work can be done by someone with technical skills who has the
hardware and the need to do it.



Re: Dell R930 server

2016-11-09 Thread Luis Coronado
the only question I see is the question about the questions.

-l

On Wed, Nov 9, 2016 at 4:42 PM, Friedrich Locke 
wrote:

> What about my questions ?
>
> Thanks.



Dell R930 server

2016-11-09 Thread Friedrich Locke
What about my questions ?

Thanks.



Re: Dell R930 server

2016-11-09 Thread Ax0n
"Nobody in their right mind would use OpenBSD for that."

That's how literally all of the projects I've used OpenBSD for have started.

On Nov 9, 2016 2:39 AM, "Martin Schröder"  wrote:

> 2016-11-09 9:06 GMT+01:00 ludovic coues :
> > I would say big data.
> >
> > Stackexchange have a pair of SQL Server, with 384Go of memory for
> > stackoverflow and 768 for everything else, a Redis server with 256, a
> > server for elasticsearch with 192 and same quantity for an HAProxy
> > server.
>
> None of this is the domain of OpenBSD and nobody in his right mind
> wants to run Stackexchange on OpenBSD.
>
> Or are you suggesting that SAP should port HANA to OpenBSD?
>
> Best
>Martin



Re: Dell R930 server

2016-11-09 Thread Martin Schröder
2016-11-09 9:06 GMT+01:00 ludovic coues :
> I would say big data.
>
> Stackexchange have a pair of SQL Server, with 384Go of memory for
> stackoverflow and 768 for everything else, a Redis server with 256, a
> server for elasticsearch with 192 and same quantity for an HAProxy
> server.

None of this is the domain of OpenBSD and nobody in his right mind
wants to run Stackexchange on OpenBSD.

Or are you suggesting that SAP should port HANA to OpenBSD?

Best
   Martin



Re: Dell R930 server

2016-11-09 Thread ludovic coues
I would say big data.

Stackexchange have a pair of SQL Server, with 384Go of memory for
stackoverflow and 768 for everything else, a Redis server with 256, a
server for elasticsearch with 192 and same quantity for an HAProxy
server.

And that's just a successful website. They aren't a search engine or a
social network

2016-11-08 23:38 GMT+01:00 Mihai Popescu :
> | Does OBSD "see" all the 96*128G memory available ?
>
> Out of curiosity, what does need such a memory today? Do you want to
> use a ramdisk?
>
> Thanks.
>



-- 

Cordialement, Coues Ludovic
+336 148 743 42



Re: Dell R930 server

2016-11-08 Thread Mihai Popescu
| Does OBSD "see" all the 96*128G memory available ?

Out of curiosity, what does need such a memory today? Do you want to
use a ramdisk?

Thanks.



Re: Dell R930 server

2016-11-06 Thread Philip Guenther
On Sun, Nov 6, 2016 at 6:11 PM, STeve Andre'  wrote:
> On 11/06/16 20:35, Philip Guenther wrote:
>>
>> On Sun, Nov 6, 2016 at 4:42 PM, Friedrich Locke
>>  wrote:
>> ...
>>>
>>> Does OBSD "see" all the 96*128G memory available ?
>>
>>
>> We only allocate a single PML4 slot for the direct map on amd64, so
>> it's currently limited to seeing 2^39 == 512GB.
>>
>> To expand that, the size and base-slot/address of the direct map
>> really need to be made variable, based on the number of physical
>> address bits supported by the CPU (as found by CPUID), preferably then
>> clamped by the range of the actual memory installed, and then set up
>> in locore.S and pmap.c
>
> Thanks for the explanation of the memory limit.  I'm not needing a
> system with more than 512G yet, but how much of a project would it
> be to dynamically expand to whatever?

Meh, not large.  Simply doubling the space supported to 1TB would be
an easy first step and -- given someone with a machine to test it --
should work out any kinks in the use of the direct map.  40 physical
bits in the professed default for long mode, so it's really what the
default should be.  Doing the second step to full dynamic would then
be about getting the ASM right when values are not longer constant.


Philip Guenther



Re: Dell R930 server

2016-11-06 Thread STeve Andre'

On 11/06/16 20:35, Philip Guenther wrote:

On Sun, Nov 6, 2016 at 4:42 PM, Friedrich Locke
 wrote:
...

Does OBSD "see" all the 96*128G memory available ?


We only allocate a single PML4 slot for the direct map on amd64, so
it's currently limited to seeing 2^39 == 512GB.

To expand that, the size and base-slot/address of the direct map
really need to be made variable, based on the number of physical
address bits supported by the CPU (as found by CPUID), preferably then
clamped by the range of the actual memory installed, and then set up
in locore.S and pmap.c


Philip Guenther




Thanks for the explanation of the memory limit.  I'm not needing a
system with more than 512G yet, but how much of a project would it
be to dynamically expand to whatever?

--STeve Andre'



Re: Dell R930 server

2016-11-06 Thread Philip Guenther
On Sun, Nov 6, 2016 at 4:42 PM, Friedrich Locke
 wrote:
...
> Does OBSD "see" all the 96*128G memory available ?

We only allocate a single PML4 slot for the direct map on amd64, so
it's currently limited to seeing 2^39 == 512GB.

To expand that, the size and base-slot/address of the direct map
really need to be made variable, based on the number of physical
address bits supported by the CPU (as found by CPUID), preferably then
clamped by the range of the actual memory installed, and then set up
in locore.S and pmap.c


Philip Guenther



Dell R930 server

2016-11-06 Thread Friedrich Locke
Does anybody here run openbsd on this machine ?
Does OBSD "see" all the 96*128G memory available ?

Thanks in advance.