Re: diglib: Meeting minutes

[JQ Johnson <jqj@darkwing.uoregon.edu>]

My understanding mirrors Carol's on most of the points, except perhaps the 
centralization issue.  I'd word the centralization consensus being that 
THIS particular proposal assumes centralization, and that when additional 
distributed storage options are needed they will probably involve different 
sets of tradeoffs and different solutions that we are not addressing at 
this time.

Centralization, by the way, means both physical placement and, more 
important, consistency in design and management.  At an extreme it could 
mean a hardware cookiecutter as we add units , but at least it means that 
we'll add units that all run pretty much the same version of Linux and all 
are managed the same way, backed up the same way on a consistent schedule, 
protected by the same type of UPS, have similar hardware architectures, 
etc.  To the extent that Travis can flesh out the details of this planned 
consistency it will assist long-term expansion planning.  Another aspect of 
the consistency is that the Linux will be pretty much the same as (or a 
subset of) that run on our many other Linux servers.

I'd also add a consensus agreement as to the importance of offline and 
offsite (tape) backup as the primary long-term archival storage, with the 
rotating storage being the working archival copy if you will.  Given the (I 
think correct) tradeoffs Travis has made between cost and robustness, I 
think it's pretty likely that we'll lose (some/much of) the data on disk at 
least once during the life of the system, and will need to depend on tape. 
So we have to have that right.