[Quantian-general] VMware clusters

[Ben Goodrich]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi Dirk,

One separate thing: Did you see the idea Fabian posted on Klaus' list a
few days ago about using a Knoppix-derived DVD as a local
apt-repository? I inferred that it would be possible for people who
didn't have much experience with Linux to update all their programs
every time a new Quantian came out without ever risking a dist-update.

Dirk Eddelbuettel wrote:
> | So, you can create a virtual machine cluster within your own computer,
> | which is probably not so useful unless you are just debugging something,
> | or you can create a (somewhat performance-hampered) virtual machine
> | cluster across computers without rebooting them out of Windows as long
> | as all the computers have VMware (especially the free-as-in-beer VMware
> | Server, which went 1.0 a few days ago).
> 
> Have you found this to be 'stable' ?  When I did some initial explorations
> with this on my small home network, I found that the 'link' between the
> virtual (inside VMware) and real machines dropped, came back, dropped, ...
> Certainly no serious test on my part but as I have not yet had a chance to
> revisit this, has anyone else?
>  

I think we are talking about slightly different things here. In general,
I get somewhat inconsistent network connectivity in the virtual machine.
So, it wouldn't surprise me if a master virtual machine would have
trouble communicating with an army of real nodes. But I haven't actually
tested this yet because the only real machines I have access to at the
moment are in an office and I wouldn't want to experiment on the office
network.

I was originally referring to a design where there was a master virtual
machine and an army of virtual nodes (preferably within separate real
machines but not necessarily). It didn't have overwhelming performance,
but I figured that was because they were all sharing a single processor
and a single pool of RAM among themselves and with the host OS. A
network of virtual machines would allow you to run something for days in
a computer lab without interfering with the ability of other people to
use the machines to check their email, etc.

> | All this appears to work with "bridged" networking and not (yet?) with
> | NAT. But it is probably smarter to set up one of those virtual network
> | things, which I haven't exactly figured out yet.
> 
> Thanks for the heads-up on this!
> 
> Cheers, Dirk
> 

My hypothesis is that if one successfully set up one of these private
networks for all the virtual machines to join, then all the load
balancing directives, etc. would be sent through a dedicated channel and
the connectivity would probably be better, too.

Ben

- --
 Ben Goodrich
 Ph.D. Student in Government and Social Policy
 http://www.people.fas.harvard.edu/~goodrich/
 goodrich at fas.harvard.edu

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.3 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFEvf/RzQDSXIcN85kRAkV8AJ9o7bZ3zX/wmavRo+h2bl1HsGZGXgCfeqxb
KoO3jWcfn4iQwxUxVXGVEE0=
=in5S
-----END PGP SIGNATURE-----