On Tue, 25 Feb 2003, Aaron Bannert wrote: > > On Monday, February 24, 2003, at 01:30 PM, Joshua Slive wrote: > >> frequency: *high* > >> small sites 4x / hour, bigger ones 1x/hour > > > > That sounds a little more agressive than is really needed. Is there a > > reason you are updating more than once a day? > > I don't think updating every hour is a problem, rsync is > pretty efficient. We shouldn't encourage more than once > and hour though.
it is a scaling issue though. maybe one rsync server syncing once an hour from the master is fine. but 100 ? rsync has a steep initial cpu usage and a reasonable memory footprint if there are a lot of files. it also will trash the namei cache (but nothing can be done about that anyway) as it traverses the filesystem area to rsync.. so doing this all hourly is probably not good. unless the apache group was releasing software hourly (a vision of coders as guinea pigs in little wheels, running harder comes to mind) i would say a daily, or at most twice daily sync should be fine. even better, it'd be good to setup a push trigger so that someone at the apache group and trigger a "push signal" which causes the mirrors to fetch. it is even nicer do this as a push signal because then the timing can be handled at the apache master server (e.g they can insert delays to stagger updates) all that needs to happen for a push trigger is adding a public key from the 'push' account on the apache server and setting it up to do the rsync fetch. regards, -jason
