[Planetlab-devel] nodes with small capacities - memory footprint, xmlrpc, and requirements on encryption s/w

[Thierry Parmentelat]

Hi folks

I am forwarding below a message from Luigi Rizzo, who wrote dummynet  
and who is contributing the integration into onelab.

My own comments are that he is one of the first ones to actually deal  
with small-capacity nodes in real, and that he is raising a very good  
point; in particular this might need to be considered in the  
discussion about using UML, or generally more strongly typed data  
models, rather than raw xmlrpc; at least in the case where one of the  
endpoints is a node, as it can be argued that infrastructures wont  
have such limitations.

The team's feelings about using variants to sha will be welcome as well.


I am attaching Luigi's code for your amusement, as he put it :)




Begin forwarded message:

> From: Luigi Rizzo <rizzo at iet.unipi.it>
> Date: May 29, 2008 11:23:26 AM CEDT
> To: Thierry Parmentelat <Thierry.Parmentelat at sophia.inria.fr>, Marta  
> Carbone <marta at prova.iet.unipi.it>
> Cc: Luigi Rizzo <rizzo at iet.unipi.it>
> Subject: xmlrpc access to the PLC
>
> Thierry,
> in doing the dummynetbox we realized that using Python to access the
> PLC requires us to load between 15 and 30MB of stuff on the image,
> just because of the tangling between the various python libraries
> that do everything.
>
> This is disturbing because the dummynet box image has limited size,
> and when it grows too much we start hitting various system limitations
> (memory disk size, etc.) which are hard to track.
>
> We are trying to replace the Python client code (NodeBootCall.py
> and NodeSessionCall.py) with some other solution, also based
> on XMLRPC of course, with a more limited memory footprint.
>
> This will also be useful in case we might want to add to the Onelab
> infrastructure some 'small' devices (e.g. shapers, routers, wireless
> gateways) which might need to talk to the PLC.
>
> At the moment we have developed some trivial C code which,
> with just 20Kbytes instead of 15MB, can run
> XMLRPC requests over https and parse the results.
>
> So far all fine, but we now need to know what kind of libraries we
> might need on the client side to perform the various processing.
>
> Right now the only obvious thing is the code used in the hmac library
>
> 	hmac.new(nodeKey, msg.encode('utf-8'), sha).hexdigest()
>
> Do you foresee any change in this area e.g. using some other
> hashing algorithm other than sha, etc. ?
>
> 	cheers
> 	luigi

-------------- next part --------------
Skipped content of type multipart/mixed