Re: Introduction to Linux based stubdom (GSoC 2011)

Re: Introduction to Linux based stubdom (GSoC 2011)

[Steven Harp]

On 05/05/2011 11:31 AM, Jiageng Yu wrote:
>
> 
>     Stubdoms are very small Xen PV guests used to run some software
> components that otherwise live in dom0. We plan to implement a linux
> based stubdom, which provides devices emulation to a particular HVM
> guest. The major tasks of this project are described below.
> 
>      1. We have to establish the environment of linux based stubdom
> with minimal linux kernel and shared libraries, which could contain
> the upstream qemu exactly. We also maintain a minimal upstream qemu by
> cutting unnecessary objs in its configuration phase. The minimal linux
> kernel, shared libraries and the upstream qemu are packed into the
> ramdisk, which is the real body of stubdom we need.
> 
...
>     This is a brief introduction to the project. Any suggestion or
> question will be Thankful!
> 

This is an excellent project.

One suggestion regarding the minimal Linux is to make allowances
in the build process for different kernel compilation options and
supplemental files in the stubdom image.   Possibly worth thinking
about different options for ramdisk, e.g. ramfs versus a compressed
ext* filesystem. The former is lighter weight, but the latter
offers more control, e.g. extended attributes.

Best of luck,
 Steve Harp

Re: Introduction to Linux based stubdom (GSoC 2011)

[Jiageng Yu]

On Sat, May 7, 2011 at 12:51 AM, Steven Harp
<steven.harp@adventiumlabs.org> wrote:
>
>
> On 05/05/2011 11:31 AM, Jiageng Yu wrote:
>>
>>
>>     Stubdoms are very small Xen PV guests used to run some software
>> components that otherwise live in dom0. We plan to implement a linux
>> based stubdom, which provides devices emulation to a particular HVM
>> guest. The major tasks of this project are described below.
>>
>>      1. We have to establish the environment of linux based stubdom
>> with minimal linux kernel and shared libraries, which could contain
>> the upstream qemu exactly. We also maintain a minimal upstream qemu by
>> cutting unnecessary objs in its configuration phase. The minimal linux
>> kernel, shared libraries and the upstream qemu are packed into the
>> ramdisk, which is the real body of stubdom we need.
>>
> ...
>>     This is a brief introduction to the project. Any suggestion or
>> question will be Thankful!
>>
>
> This is an excellent project.
>
> One suggestion regarding the minimal Linux is to make allowances
> in the build process for different kernel compilation options and
> supplemental files in the stubdom image.   Possibly worth thinking
> about different options for ramdisk, e.g. ramfs versus a compressed
> ext* filesystem. The former is lighter weight, but the latter
> offers more control, e.g. extended attributes.
>
> Best of luck,
>  Steve Harp
>
>
>
>

Hi Steven,

    Thanks very much for the advice. Honestly, until now, I have no
idea about the performance of Linux based stubdoms. So I try to make
it minimal first. Additionally, though a compressed ext* filesystem
could offer more control, stubdoms have little chance of accessing the
local filesystem in our current implementment plan.

    I will confirm with my mentor whether stubdoms need more
functional requirements. Thank you a lot.

    Regards!

Jiageng Yu.

Introduction to Linux based stubdom (GSoC 2011)

[于佳耕]

Hi all,

    I am Jiageng Yu. I come from Institute of Software, Chinese
Academy of Sciences, Beijing, China. I have been accepted to GSoC 2011
and will work on the project of Linux based stubdom in next three
monthes. I am greatly honored and excited to join the Xen community.

    Stubdoms are very small Xen PV guests used to run some software
components that otherwise live in dom0. We plan to implement a linux
based stubdom, which provides devices emulation to a particular HVM
guest. The major tasks of this project are described below.

     1. We have to establish the environment of linux based stubdom
with minimal linux kernel and shared libraries, which could contain
the upstream qemu exactly. We also maintain a minimal upstream qemu by
cutting unnecessary objs in its configuration phase. The minimal linux
kernel, shared libraries and the upstream qemu are packed into the
ramdisk, which is the real body of stubdom we need.

     2. We could adapt simulation devices of upstream qemu to the
relevant frontend devices of stubdom in optimal ways. In this
scenario, we will consider the devices including network, disk and
graphics.

     3. We could also implement the proper save/restore mechanism of
simulation devices. As we all know, qemu running in the stubdom writes
a statefile that has to be returned to the toolstack in dom0.
Currently minios based stubdoms use two PV consoles to do that, which
have not been supported in Linux. Therefore, instead of additional
consoles, we create a dynamic virtual disk to transfer save/restore
data between Dom0 and stubdom.

     4. We plan to make stubdom to support pci pass-through devices.
Our goal is to map the memories (including ioport, iomem and irq) of
pass-through devices into memories of frontend devices in stubdom.
That could be a few differences to the implementation of MiniOS based
stubdom.

     5. At last, we produce benchmarks to measure the resource usage
and the level of performances of linux stubdom in comparison with
MiniOS stubdom. We will measure the throughputs such as CPU, disk and
network, etc.

    This is a brief introduction to the project. Any suggestion or
question will be Thankful!

    Best Regards!


Jiageng Yu

Re: Introduction to Linux based stubdom (GSoC 2011)

[Samuel Thibault]

Hello,

于佳耕, le Thu 05 May 2011 23:45:29 +0800, a écrit :
>     I am Jiageng Yu. I come from Institute of Software, Chinese
> Academy of Sciences, Beijing, China.

Welcome!

>      5. At last, we produce benchmarks to measure the resource usage
> and the level of performances of linux stubdom in comparison with
> MiniOS stubdom. We will measure the throughputs such as CPU, disk and
> network, etc.

That will be of much interest to me indeed: maintainability vs
performance :)

Please explicitly Cc me when posting patches/comments/etc. to the
mailing list, to catch my attention.  Also do not hesitate to Cc me when
asking questions.  I didn't know exactly about the Xen GSoC projects so
didn't join for mentoring, but I can help.

Samuel

Re: Introduction to Linux based stubdom (GSoC 2011)

[Jiageng Yu]

On Fri, May 6, 2011 at 12:12 AM, Samuel Thibault
<samuel.thibault@ens-lyon.org> wrote:
> Hello,
>
> 于佳耕, le Thu 05 May 2011 23:45:29 +0800, a écrit :
>>     I am Jiageng Yu. I come from Institute of Software, Chinese
>> Academy of Sciences, Beijing, China.
>
> Welcome!
>
>>      5. At last, we produce benchmarks to measure the resource usage
>> and the level of performances of linux stubdom in comparison with
>> MiniOS stubdom. We will measure the throughputs such as CPU, disk and
>> network, etc.
>
> That will be of much interest to me indeed: maintainability vs
> performance :)
>
> Please explicitly Cc me when posting patches/comments/etc. to the
> mailing list, to catch my attention.  Also do not hesitate to Cc me when
> asking questions.  I didn't know exactly about the Xen GSoC projects so
> didn't join for mentoring, but I can help.
>
> Samuel
>

I will do it. Thanks so much!