At 01:17 AM 22/09/2004 -0500, Ivan R. Judson wrote:
>I would love to see this discuss very broadly so we can come up with a set
>of requirements to solve this problem once. 

Sounds good. I need to take a raincheck on this topic, as I'm about to hit the 
road for a week, with very limited email access, but will write some thoughts 
for yelling at when I get back :-)

>Here are the considerations that I think need to be remembered:
>
>1) How do you do this kind of thing generically (ie, what if the software
>needs user input, has a UI, etc)
>2) How do you limit the execution to trusted code? or otherwise limit the
>liability of potentially running arbitrary code on any machine in your node?
>3) Can these apps/services be launched or controlled by users elsewhere?
>4) What's the minimum interface that is needed to explore these issues?

Slightly ironically, these are the same issues that the computational grid has 
to deal with. However, one thing I would suggest is that there is probably a 
spectrum of solutions - we can solve the general case, or we can solve specific 
cases, or a point in between. The specific cases are almost certainly easier. 
e.g. keeping it to apps that we actually install manually (like shared app 
packages that we trust), that can only run and be controlled within the sandbox 
of a node's set of machines, etc. Ultimately, the general case would allow us 
to tie the AG into computational steering and data grids and distributed 
visualisation render-farms and so forth, and that is somewhere we want to be as 
well. But we also want to solve the short term problem of having ppt start up 
on the machine *next* to the venue client machine. But all of the issues you 
mention (and others) are very important, and need to be addressed, at either 
end of the scale. Trying a few approaches might be better!
 t!
han 
building only from the requirements down?

Anyway, it's late, and I don't have Ivan's sleep patterns. I hope there's a 
long thread for me to read when I get back online :-) or else I'll suggest some 
really stupid things then ;-) (again? Who said 'again'?)

Cheers,
        Markus



>--Ivan 
>
>> -----Original Message-----
>> From: owner-ag-t...@mcs.anl.gov 
>> [mailto:owner-ag-t...@mcs.anl.gov] On Behalf Of Markus Buchhorn
>> Sent: Tuesday, September 21, 2004 11:07 PM
>> To: ag-t...@mcs.anl.gov
>> Subject: [AG-TECH] SharedApp on non-venue-client machine?
>> 
>> Hi All
>> 
>> I'm looking at better ways of managing our (multi-machine) 
>> nodes. One of the things I want to do is have the operation 
>> be remote to the node, by having the venue-client running on 
>> (for example) the control machine. (I'd really like to have 
>> multiple venue clients (or just one?) on a single machine, 
>> operating multiple separate AG nodes across our campuses, but 
>> that's a little way off yet).
>> 
>> The one thing that gets in the way seems to be that 
>> SharedApps (browser, presentations, etc.) appear on the same 
>> machine as the venue client. This seems to tie the 
>> SharedApps, the venue client and the display machine 
>> together. That seems to be against the architecture of 
>> distributed node-services and service managers. So, I'm 
>> probably doing something wrong :-)
>> 
>> Is there a way of configuring a node so that the sharedapps 
>> appear on the display machine (say), rather than the 
>> venue-client machine? 
>> 
>> Thanks!
>> 
>> Cheers,
>>       Markus
>> 
>> Markus Buchhorn, ANU Internet Futures             |Ph: +61 2 61258810
>> markus.buchh...@anu.edu.au                        |Fx: +61 2 61259805
>> The Australian National University, Canberra 0200 |Mob: 0417 281429
>> 
>> 


Markus Buchhorn, ANU Internet Futures             |Ph: +61 2 61258810
markus.buchh...@anu.edu.au                        |Fx: +61 2 61259805
The Australian National University, Canberra 0200 |Mob: 0417 281429

Reply via email to