[Swift-devel] oops provenance

[Michael Wilde]

was: Re: first stab

Glen, I'd like to pick up on the comment on provenance you made last 
Sunday while gathering data for the oops paper.

Ben is focusing on provenance at the moment, and so I'd like to include 
him in the discussion (although I know you are focusing on the 
provenence challenge at the moment).

I think a starting point for oops provenance is this: For every run, you 
want to know:

- an ID for the run
- time and date started / ended
- how many jobs ran where
- location of the output files & logs
- the run parameters (proteins, config params, scale)
- analyzed scores of the run output
- what version of oops was used
- what version of swift/cog was used
- what version of the oops.swift script was used

Given this in a database, you could also compare structure scores for 
one version of code or one algorithm vs another

We're also always looking to see what level of parallelism was achieved 
by swift, so some way of getting that out of the logs, up to an 
including full log plots, would be handy.

- Mike



On 4/12/09 5:51 PM, Glen Hocky wrote:
> well, if i'd done a summary before hand, i may have tried to do a few 
> extra proteins or something. anyway, i think everything is fine, but i 
> am definitely going to have to think about some way of summarizing. i 
> think this makes a good case for provenance tracking though :)
> 
> Michael Wilde wrote:
>> Hi Glen,
>>
>> Not sure what you mean by "... annoyed at the bredth of the runs that 
>> i've done. i may do a few more on abe qb and  ranger if they are 
>> working because that would just take a few hours"
>>
>> As in the runs are not yielding something interesting to write about?
>>
>> One thing we can talk about is just the run time, etc. We havent 
>> looked at that closely, but hopefully its good enough to be worth citing.
>>
>> Anything I can do to help organize or discuss this section with you?
>>
>> - Mike