Re: [Tzdist] Fwd: [tzdist] #32 (service): managing historical data

[Paul Eggert <[e-mail]>]

Doug Royer wrote:
> The update might change 10 times a year. I would think the CUAs will be checking
> a lot more often

If so, the overall cost of checking-for-no-change will completely swamp the 
overall cost of transferring data (on the rare occasions where an update is 
actually needed).  So why expend design and implementation effort optimizing a 
case that doesn't much matter?

I'd feel better about this if there were some analysis justifying why the 
protocol has relatively-complicated features to support incremental updates. 
The back-of-the-envelope calculations I've done make it look like they're more 
trouble than they're worth.