bash or any other $favourite?

On Sun, Jun 11, 2006 at 06:09:38PM +0000, Peter Corlett
wrote:
> Chris Benson <chrisbjesmond.demon.co.uk>
wrote:
> [...]
> > Ah, that may reduce my interest in UML before the
book even arrives  I
> > was hooked by the promise of UML sharing the
host's o/s image, but that
> > uses Copy-on-Write ... so I'm unclear whether
local changes in the child
> > will be visible to the host.
> 
> Well, it is possible that UML does have the capability,
but it seems
> unlikely. I have no experience of it to say one way or
the other.
> 
> Basically, virtualisation falls into two camps:
> 
> a) chroot-on-speed, e.g. linux-vserver, FreeBSD jails,
Solaris zones.
>    Everything runs under the same kernel instance.
It's a very lightweight
>    scheme. Linux-vserver suffers from some interesting
gotchas due to the
>    single kernel instance, and I imagine the others are
similar.
> 
>    Due to the single kernel instance, there's only one
thing accessing the
>    filesystem so the controlling/root server can access
the filesystems.
> 
> b) Individual kernel instances that run separately,
e.g. UML, Xen, VMWare.
>    Everything looks and acts like a physical machine.
It's relatively
>    heavyweight but the extra separation can make this
worthwhile.
> 
>    Because each kernel has its own buffer cache et al,
it's impractical to
>    mount filesystems in use by another virtual server
except over something
>    like NFS.

Although lvm snapshots might be useful in such a context. 

The real question is what do you want the visibility for ?

Clustered fileystems and nbds, like drbd, also come to mind,
but I can't
immediately imagine why choose them over the chroot-on-speed
option, but 
it comes to mind in the context of either test or vm
migration.

Regards,
Paddy
-- 
Perl 6 will give you the big knob. -- Larry Wall

On Mon, 2006-06-12 at 10:05 +0100, paddy wrote:

> Clustered fileystems and nbds, like drbd, also come to
mind, but I can't
> immediately imagine why choose them over the
chroot-on-speed option, but 
> it comes to mind in the context of either test or vm
migration.
> 

Ah, glad you mentioned clustered file systems: what is the
clustered
file system du jour for Linux at the moment?

Dirk

On Mon, Jun 12, 2006 at 11:47:08AM +0100, Dirk Koopman
wrote:

> Ah, glad you mentioned clustered file systems: what is
the clustered
> file system du jour for Linux at the moment?

Do you mean the popular and fashionable one that people are
using
currently as their preference because it is trendy and in
vogue right
now?

-- 
Paul Johnson - paulpjcj.net
http://www.pjcj.net

On Mon, Jun 12, 2006 at 11:47:08AM +0100, Dirk Koopman
wrote:
> On Mon, 2006-06-12 at 10:05 +0100, paddy wrote:
> 
> > Clustered fileystems and nbds, like drbd, also
come to mind, but I can't
> > immediately imagine why choose them over the
chroot-on-speed option, but 
> > it comes to mind in the context of either test or
vm migration.
> > 
> 
> Ah, glad you mentioned clustered file systems: what is
the clustered
> file system du jour for Linux at the moment?

not my area, and I think it depends on what your are trying
to do, but 
I'm vaguely aware of stuff like OCFS2, GFS, lustre ...

I think at least OCFS and GFS provide the kind of shared
access that would 
provide the similar to nfs kind of simultaneous access that
I was suggesting.
Of course, arguably they're just another form of network
filesystem, but 
I wanted to mention lvm snapshots, and they came to mind too


I guess I'm also thinking about the contrast between a xen
style system,
and a more hurd-like idea of distinct filesystem servers. 
The question
being, why do we we view the "then you'd have to run
nfs" as somehow
strange, and the answer likely being performance (and
perhaps history).

Regards,
Paddy
-- 
Perl 6 will give you the big knob. -- Larry Wall

On Mon, Jun 12, 2006 at 01:08:24PM +0200, Paul Johnson
wrote:
> On Mon, Jun 12, 2006 at 11:47:08AM +0100, Dirk Koopman
wrote:
> 
> > Ah, glad you mentioned clustered file systems:
what is the clustered
> > file system du jour for Linux at the moment?
> 
> Do you mean the popular and fashionable one that people
are using
> currently as their preference because it is trendy and
in vogue right
> now?

LOL

OK, digging the hole I'm in much deeper ...

I was thinking it might possible to serve up some shared
block device
abstraction, such as I think the GFS authors were playing
with, and
get an elegant, perhaps even high performance, result 

Regards,
Paddy
-- 
Perl 6 will give you the big knob. -- Larry Wall

On Mon, 2006-06-12 at 13:08 +0200, Paul Johnson wrote:
> On Mon, Jun 12, 2006 at 11:47:08AM +0100, Dirk Koopman
wrote:
> 
> > Ah, glad you mentioned clustered file systems:
what is the clustered
> > file system du jour for Linux at the moment?
> 
> Do you mean the popular and fashionable one that people
are using
> currently as their preference because it is trendy and
in vogue right
> now?
> 

Probably not, but it could be, if it is any good.

I suppose what I am looking for is some kind of RAID 5
effort, but
distributed across a few machines.

On Mon, Jun 12, 2006 at 03:01:51PM +0100, Dirk Koopman
wrote:
> 
> I suppose what I am looking for is some kind of RAID 5
effort, but
> distributed across a few machines.

Is that what GoogleFS was about?  There were several
write-ups a few weeks 
ago.  Of course it helps to have a few thousand white-box
PCs with your 
own o/s tweaks applied ...

On 12 Jun 2006, at 15:01, Dirk Koopman wrote:

> On Mon, 2006-06-12 at 13:08 +0200, Paul Johnson wrote:
>> On Mon, Jun 12, 2006 at 11:47:08AM +0100, Dirk
Koopman wrote:
>>
>>> Ah, glad you mentioned clustered file systems:
what is the clustered
>>> file system du jour for Linux at the moment?
>>
>> Do you mean the popular and fashionable one that
people are using
>> currently as their preference because it is trendy
and in vogue right
>> now?
>>
>
> Probably not, but it could be, if it is any good.
>
> I suppose what I am looking for is some kind of RAID 5
effort, but
> distributed across a few machines.

gigabit, NFS over RAID 5?

-- 
Dave Hodgkinson - Music photography
http://www.davehodgkin
son.com/

On 12 Jun 2006, at 15:10, Chris Benson wrote:

> On Mon, Jun 12, 2006 at 03:01:51PM +0100, Dirk Koopman
wrote:
>>
>> I suppose what I am looking for is some kind of
RAID 5 effort, but
>> distributed across a few machines.
>
> Is that what GoogleFS was about?  There were several
write-ups a  
> few weeks
> ago.  Of course it helps to have a few thousand
white-box PCs with  
> your
> own o/s tweaks applied ...

GoogleFS is very special and very geared to what they do.
The white  
paper
is fascinating.


-- 
Dave Hodgkinson - Music photography
http://www.davehodgkin
son.com/

On Mon, Jun 12, 2006 at 03:01:51PM +0100, Dirk Koopman
wrote:
> On Mon, 2006-06-12 at 13:08 +0200, Paul Johnson wrote:
> > On Mon, Jun 12, 2006 at 11:47:08AM +0100, Dirk
Koopman wrote:
> > 
> > > Ah, glad you mentioned clustered file
systems: what is the clustered
> > > file system du jour for Linux at the moment?
> > 
> > Do you mean the popular and fashionable one that
people are using
> > currently as their preference because it is trendy
and in vogue right
> > now?
> > 
> 
> Probably not, but it could be, if it is any good.
> 
> I suppose what I am looking for is some kind of RAID 5
effort, but
> distributed across a few machines.

Is slashdot still "trendy and in vogue" ?

http://hardware.slashdot.org/hardware/06/06/14/03262
21.shtml

Regards,
Paddy
-- 
Perl 6 will give you the big knob. -- Larry Wall