[Qemu-devel] guest kernel in the host RAM

[Qemu-devel] guest kernel in the host RAM

[Denis V. Lunev]

Hello, Paolo!

There is a plan or idea to improve current implementation
of Clear Containers for QEMU. Patches about NVDIMM
are floating in the list, you have proposed special
boot firmware for a case.

The idea is that we could boot with the uncompressed Linux
kernel in the guest which serves 2 purposes:
- faster boot time as per Intel opinion
- memory savings

Uncompressed kernel image could be the same for several
guests and thus memory pages for the kernel could be the
same, which saves a bit of RAM :)

At the moment neither BIOS support this in full: nor SeaBIOS,
nor OVMF, nor QBoot.

 From our point of view QBoot is an ideal candidate for this.

Do you have any opinion for the topic?

Den

Re: [Qemu-devel] guest kernel in the host RAM

[Paolo Bonzini]

On 15/12/2015 12:27, Denis V. Lunev wrote:
> The idea is that we could boot with the uncompressed Linux
> kernel in the guest which serves 2 purposes:
> - faster boot time as per Intel opinion
> - memory savings
> 
> Uncompressed kernel image could be the same for several
> guests and thus memory pages for the kernel could be the
> same, which saves a bit of RAM :)
> 
> At the moment neither BIOS support this in full: nor SeaBIOS,
> nor OVMF, nor QBoot.
> 
> From our point of view QBoot is an ideal candidate for this.

QBoot is not meant for use in production; SeaBIOS actually is just as 
fast if you configure it right.

> Do you have any opinion for the topic?

I honestly don't like the idea.  Linux patches itself a lot, which 
makes the memory savings minimal (if they exist at all).  What is the
decompression time for a kernel that is compressed with LZO?

Paolo

Re: [Qemu-devel] guest kernel in the host RAM

[Stefan Hajnoczi]

[-- Attachment #1: Type: text/plain, Size: 462 bytes --]

On Tue, Dec 15, 2015 at 02:27:14PM +0300, Denis V. Lunev wrote:
> There is a plan or idea to improve current implementation
> of Clear Containers for QEMU. Patches about NVDIMM
> are floating in the list, you have proposed special
> boot firmware for a case.

By the way, Marc Mari is working on adding fw_cfg DMA support to
linuxboot.bin in QEMU.  That way kernel/initramfs loading is much
quicker because it doesn't use PIO to read data from the host.

Stefan

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 473 bytes --]

Re: [Qemu-devel] guest kernel in the host RAM

[Miao Yan]

2015-12-15 19:27 GMT+08:00 Denis V. Lunev <den@openvz.org>:
> Hello, Paolo!
>
> There is a plan or idea to improve current implementation
> of Clear Containers for QEMU. Patches about NVDIMM
> are floating in the list, you have proposed special
> boot firmware for a case.
>
> The idea is that we could boot with the uncompressed Linux
> kernel in the guest which serves 2 purposes:
> - faster boot time as per Intel opinion
> - memory savings
>
> Uncompressed kernel image could be the same for several
> guests and thus memory pages for the kernel could be the
> same, which saves a bit of RAM :)
>
> At the moment neither BIOS support this in full: nor SeaBIOS,
> nor OVMF, nor QBoot.

Just FYI, U-boot can load uncompressed kernels and it can
run on qemu.

>
> From our point of view QBoot is an ideal candidate for this.
>
> Do you have any opinion for the topic?
>
> Den
>

Re: [Qemu-devel] guest kernel in the host RAM

[Denis V. Lunev]

On 12/15/2015 02:59 PM, Paolo Bonzini wrote:
>
> On 15/12/2015 12:27, Denis V. Lunev wrote:
>> The idea is that we could boot with the uncompressed Linux
>> kernel in the guest which serves 2 purposes:
>> - faster boot time as per Intel opinion
>> - memory savings
>>
>> Uncompressed kernel image could be the same for several
>> guests and thus memory pages for the kernel could be the
>> same, which saves a bit of RAM :)
>>
>> At the moment neither BIOS support this in full: nor SeaBIOS,
>> nor OVMF, nor QBoot.
>>
>>  From our point of view QBoot is an ideal candidate for this.
> QBoot is not meant for use in production; SeaBIOS actually is just as
> fast if you configure it right.
>
>> Do you have any opinion for the topic?
> I honestly don't like the idea.  Linux patches itself a lot, which
> makes the memory savings minimal (if they exist at all).  What is the
> decompression time for a kernel that is compressed with LZO?
>
> Paolo
It seems that you are perfectly correct here!

With large pages enabled (and preferred 2 Mb pages mode)
we will have exactly zero bonus.

Thank you for saving a lot of time and efforts :)))

Den

Re: [Qemu-devel] guest kernel in the host RAM

[Paolo Bonzini]

On 18/12/2015 17:28, Denis V. Lunev wrote:
>> Linux patches itself a lot, which
>> makes the memory savings minimal (if they exist at all).  What is the
>> decompression time for a kernel that is compressed with LZO?
>
> It seems that you are perfectly correct here!
> 
> With large pages enabled (and preferred 2 Mb pages mode)
> we will have exactly zero bonus.
> 
> Thank you for saving a lot of time and efforts :)))

I'm a lazy person, sometimes it helps. :)

Paolo