On Wed, May 11, 2016 at 12:26 AM, Frank Erjiang Fu <F.Fu@xxxxxxxxxx> wrote:
> Michael, Interesting to know that you have a pool of THREDDS servers. With my
> example, would you have THREDDS cluster on Thredds 3 or Thredds 1 and Thredds
> 2? I guess that it depends on what kind of queries (e.g. large number of
> small queries or small number of big queries) right?
In our case ALL the servers in the DNS-RR pool *need* to see & serve
the exact same data. The catalogs are identical (puppet controlled),
the cache/agg folder for THREDDS is even "synchronized" between
servers via a BitTorrent Sync share. The DNS-RR pooling method is not
perfect, but is the easiest to setup.
http://bit.ly/1XneWqb
The goal is to serve the same data using as many data pipes as
possible... and hope that all servers "answer" when their IP is chosen
at random from the DNS-RR pool :)
http://hycom.org/tools/status
--
Michael McDonald
Florida State University
