Re: Problems about Xen's Howto

Re: Problems about Xen's Howto

[Bin Ren]

On 2 Feb 2004, at 11:23, Yan-Ching CHU wrote:

> Hi Bin,
>  
>     I have got some problem starting new domains following the Xen 
> Howto v1.1 that I hope you can help. In both the README and your howto 
> there is no explicit statement about rather new domains can/should be 
> started using another filesystem. I wonder if I use the same root as 
> Domain0 is using to start new domains, rather the original filesystem 
> will be destroyed due to multiple writers to the filesystem.

I'm sorry I didn' t make it clear. You are absolutely right that each 
domain
should use a different root filesystem.

> If I really need new fileysystem for every new domain, what minimum 
> set of services/files about the boot process (inittab, init rc, 
> daemons) do I need to have? The whole thing in the DemoCD is quite big 
> (especially with /usr) that I don't want to install each copy of them 
> to every domain I want to start. Any ideas/comments?

One independent root filesystem for each domain requires too much space,
this is a major problem of today's Xen that we are trying to tackle. 
The method
is to use Copy on Write, e.g. different domains share the same files as 
many
as possible and retains modified files separately. CoW can be done both 
in
user space and in kernel space.

Russ has implemented a CoW NFS and I'm at the last staging of writing a
CoW multi-disk driver for linux kernel 2.4 series.

Before CoW features become stable and robust, you can:

(1) duplicate root filesystems for each domain (Yes, I can hear you 
screaming...)
(2) share /usr among all the domains to save a lot of space (Yes, let's 
relax...)
(3) get a new hard disk and go to either (1) or (2)

Thanks,
Bin


-------------------------------------------------------
The SF.Net email is sponsored by EclipseCon 2004
Premiere Conference on Open Tools Development and Integration
See the breadth of Eclipse activity. February 3-5 in Anaheim, CA.
http://www.eclipsecon.org/osdn

Questions on Copy-on-Write

[Yan-Ching CHU]

Thanks for your input Bin.

I would love to know your progress of CoW for linux kernel 2.4, will you
integrate it into Xen later? Any plan for a Linux 2.6 port? Any
papers/materials I can refer to?

Regarding the Copy-on-Write NFS is the group using it with Xen? Any one has
experience/comments of using it with Xen?

Cheers,
Yan-Ching CHU




----- Original Message ----- 
From: "Bin Ren" <br260@cam.ac.uk>
To: "Yan-Ching CHU" <cs0u210a@liverpool.ac.uk>
Cc: "Devel Xen" <xen-devel@lists.sourceforge.net>
Sent: Thursday, January 01, 1970 1:10 AM
Subject: Re: Problems about Xen's Howto


On 2 Feb 2004, at 11:23, Yan-Ching CHU wrote:

> Hi Bin,
>
> I have got some problem starting new domains following the Xen
> Howto v1.1 that I hope you can help. In both the README and your howto
> there is no explicit statement about rather new domains can/should be
> started using another filesystem. I wonder if I use the same root as
> Domain0 is using to start new domains, rather the original filesystem
> will be destroyed due to multiple writers to the filesystem.

I'm sorry I didn' t make it clear. You are absolutely right that each
domain
should use a different root filesystem.

> If I really need new fileysystem for every new domain, what minimum
> set of services/files about the boot process (inittab, init rc,
> daemons) do I need to have? The whole thing in the DemoCD is quite big
> (especially with /usr) that I don't want to install each copy of them
> to every domain I want to start. Any ideas/comments?

One independent root filesystem for each domain requires too much space,
this is a major problem of today's Xen that we are trying to tackle.
The method
is to use Copy on Write, e.g. different domains share the same files as
many
as possible and retains modified files separately. CoW can be done both
in
user space and in kernel space.

Russ has implemented a CoW NFS and I'm at the last staging of writing a
CoW multi-disk driver for linux kernel 2.4 series.

Before CoW features become stable and robust, you can:

(1) duplicate root filesystems for each domain (Yes, I can hear you
screaming...)
(2) share /usr among all the domains to save a lot of space (Yes, let's
relax...)
(3) get a new hard disk and go to either (1) or (2)

Thanks,
Bin



-------------------------------------------------------
The SF.Net email is sponsored by EclipseCon 2004
Premiere Conference on Open Tools Development and Integration
See the breadth of Eclipse activity. February 3-5 in Anaheim, CA.
http://www.eclipsecon.org/osdn

Re: Questions on Copy-on-Write

[Bin Ren]

On 2 Feb 2004, at 12:49, Yan-Ching CHU wrote:

> I would love to know your progress of CoW for linux kernel 2.4, will 
> you
> integrate it into Xen later? Any plan for a Linux 2.6 port? Any
> papers/materials I can refer to?

I'm in the very last stage of testing/debugging/enhancing/refining/
CoW for 2.4. Because it's a driver for Linux kernel, it's quite
independent from Xen and thus requires little (if any) Xen modification.
Just create a CoW multi-disk device with VBDs in XenoLinux
and that's it.

As for 2.6 port, it depends on how happy people are with this CoW
driver. I haven't written any document yet. But you may want to
take a look at Software-RAID HOWTO.

Cheers,
Bin



-------------------------------------------------------
The SF.Net email is sponsored by EclipseCon 2004
Premiere Conference on Open Tools Development and Integration
See the breadth of Eclipse activity. February 3-5 in Anaheim, CA.
http://www.eclipsecon.org/osdn

Re: Re: Questions on Copy-on-Write

[Christian Limpach]

Hi!

> I'm in the very last stage of testing/debugging/enhancing/refining/
> CoW for 2.4. Because it's a driver for Linux kernel, it's quite
> independent from Xen and thus requires little (if any) Xen modification.

Have you looked at device-mapper?  I think it's included in 2.6 and it's
also available for 2.4 and it includes a CoW mapping module.  It's really
well designed, I liked it enough to bother implementing it on NetBSD.

> Just create a CoW multi-disk device with VBDs in XenoLinux
> and that's it.

Doesn't this sit at the wrong level?  This would make all disk accesses on
these devices go through Domain0?

    christian



-------------------------------------------------------
The SF.Net email is sponsored by EclipseCon 2004
Premiere Conference on Open Tools Development and Integration
See the breadth of Eclipse activity. February 3-5 in Anaheim, CA.
http://www.eclipsecon.org/osdn

Re: Re: Questions on Copy-on-Write

[Ian Pratt]

 
> > Just create a CoW multi-disk device with VBDs in XenoLinux
> > and that's it.
> 
> Doesn't this sit at the wrong level?  

Implementing the CoW functionality in Linux is basically a
stop-gap until we can do it more cleanly with the new IO stuff
that's under development.

> This would make all disk accesses on these devices go through
> Domain0?

Any domain that wanted CoW disk access could use Bin's CoW driver
directly. Of course, that would no longer be transparent to the
guest OS. It'll all be much cleaner in the future...

Cheers,
Ian


-------------------------------------------------------
The SF.Net email is sponsored by EclipseCon 2004
Premiere Conference on Open Tools Development and Integration
See the breadth of Eclipse activity. February 3-5 in Anaheim, CA.
http://www.eclipsecon.org/osdn