paul at anastrophe.com
Sat Sep 29 04:32:28 UTC 2018
On 9/28/2018 20:56 PM, Hal Murray wrote:
> Agreed, mostly. There is a big difference between the traffic from a few
> geeks occasionally downloading a file and a swarm of systems downloading from
> a cron job.
> Years ago, a friend introduced me to a great term. I asked him what happens
> if his application took off? Did he have plans for the servers. His response
> was "That's a success disaster. If that happens, we can get the resources we
> need." But that context was all within one organization. With something like
> NIST or IETF, our success disaster could be their problem.
Perhaps, but being a pretty tiny text file, I'd imagine it would take an
awful lot to affect them - again, particularly since they already
distribute the file via ftp, which I'm sure incurs its own fair share of
traffic. Over https of course it involves a lot more horsepower per file
A crafty way to avoid the 'every x on the x' problem of cron jobs - which
can contribute to such 'success disasters' - might be to employ the
systemd timer functionality for leapfetch, e.g. the apt-daily.timer that
pushes its checks around to begin it at random times. Since the
leap-seconds file I think only needs to be checked once a month - at most
- that could ensure that even with millions of ntpsec installations, it'd
be relatively trivial traffic load, widely dispersed.
More information about the devel