Subject: Re: installing into RAID-1: HOWTO and example
To: NetBSD/xen <port-xen@NetBSD.org>
From: Johan Ihren <email@example.com>
Date: 05/17/2007 12:35:34
-----BEGIN PGP SIGNED MESSAGE-----
I wrote (about a month ago):
On 19 Apr 2007, at 22:16, Johan Ihren wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>>>> Yes, this is the PCI hole. I've got the same issue with a i945.
>>>> Despite the claim that the motherboard "supports" 4Gb and using a
>>>> EM64T CPU,
>>>> 768M were stolen for PCI. By playing with BIOS options, disabling
>>>> all unused
>>>> devices I trimmed this down to ~600M.
>>>> In order to really use 4GB you need a i955X chipset.
>>> I have an i965 chipset and I also lose 768MB of 4GB to the PCI hole
>>> (it's sort of ok, at least I was aware of it when deploying this
>> With a 32bits or 64bits OS ? with 32bits, you won't see more than 4GB
>> including the PCI hole anyway, even if the chipset supports
>> remaming the
>> remaining RAM above 4GB.
>>> However, my question is what would happen if I add more memory to
>>> machine. I.e. if I boost it to 8GB would I still only get 3.25GB for
>>> Xen or would I get 7.25GB not having to bother about the hole in the
>> NetBSD/Xen is still 32bits, so it's limited to 4GB.
> That's what I feared. Yes, I understand that a 32bit OS is limited
> to 4GB, but I thought that perhaps the Xen hypervisor had no such
> restrictions and as long as dom0 and all domU's where small enough
> it might work. But then I've also seen the discussions about PAE
> and page table structures having to be the same between hypervisor,
> dom0 and domU, so I'm not surprised, really.
> Ho hum. I suspect this will bite me hard around the end of the
> year. Oh, well.
It seems that I've already been bitten, but in a new and unexpected way.
I upgraded a system with a Q965 chipset from 2GB to 4GB fully aware
that I'd lose the last 0.75GB. However, this time I only got 2.75GB
(rather than 3.25) in the end which more or less blows my memory
budget for this server.
Apart from the known limitation that the PCI hole will steal (i.e.
make inaccessible) a large chunk of memory I find it really annoying
that the size of the chunk seems to be so unpredictable.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (Darwin)
-----END PGP SIGNATURE-----