ORBIT-USER: Node image repository?

[Max Ott]

We are persuing a different strategy. Start with a clean, well
documented image (basically the kernel version and it's configuration
file, and a list of APT packages) and a single experiment file
describing the additionally installed packages plus all the expriment
specific configurations.

This way you have everything in a single place.

The problem with images are two-fold:

* The modifications are rarely documented. So you may have a reference
to an image but you may not be entirely sure what's on there and how
it is really set-up without a lot of aditional investigative digging

* If you work with an outdoor testbed, you may not have the network
capacity to pull down large images.

We can argue about the first point, but I'm right now working with a
testbed where my control channel is a commercial WiMax with a
1GB/month limit.

I agree that hand-crafting an image is much easier than packaging all
your applications into APT files and using a setup mechanism which is
lacking capabilities. But if you get over the initial hurdle it isn't
that difficult and a systematic and well structured approach to
experiments is paying off nicely in the long run.

I do agree that we have to dramatically reduce the initial learning
curve, but we are chipping away on it. To give you an idea of where we
are heading in this regard, have a look at
http://ireel.npc.nicta.com.au/ This is right now used for simple
fixed-network experiments (and is not using Orbit) in the context of
networking courses, but it shouldn't be hard to imagine that one set's
up something similar as a companion to a paper allowing readers to
explore your results and finding further.

Regards,

-max

On 8/29/06, Einar Vollset <einar at cs.cornell.edu> wrote:
>
>
> Hi,
>
> I was wondering - how much work would it be to create a node image
> repository with a web interface? I'm thinking of something like the
> ability to describe and make publicly available a saved node image
> (note that by "making available" I don't downloadable from the
> website, but accessible from the orbit-lab consoles).
>
> This would be quite useful, not only to people like me lacking the
> mental capacity to remember what the node images are called, or
> indeed which one does what, but also as it may cut down on
> replication of work, encourage cooperation and heighten the profile
> of good, real wireless systems built.
>
> For example, if researcher A works on MAC scheduling algorithms, and
> (doing some good work) gets it published in MobiCom, he could put up
> a page describing the node image, and then reference that in his paper.
>
> Researcher B, being more interested in routing protocols, could then
> use researcher A's image to build his system over. When researcher B
> then comes to write his paper, he references researcher A's image &
> paper.
>
> Going even further you could imagine researcher C (a surly type)
> deciding that researcher A's MAC protocol is crap, and building his
> own to beat it. He then does a performance test between his new
> algorithm and researcher A's (as he easily has that available),
> publishes a paper, etc, etc..
>
> You could even give a prize to the most used node image, or most
> reference node image. ;-)
>
> Cheers,
>
> Einar
>
>
>
>
>
>
> ------------------------------------------------------
> Einar Vollset
> Visiting Assistant Professor
> Department of Computer Science
> Cornell University
>
> Phone: +1-607-255-7573
> Web: http://www.cs.cornell.edu/~einar
> ------------------------------------------------------
>
>
>
>
>