Load is ditributed on the basis of URI, with parameter set in haproxy
config as "balance uri".
This has been done to achieve maximum Cache Hit from the Server.
While the cache might be more efficient this way this can lead to one server
always serving some "hot" content while others stay idle.
If you can afford to shrink the cache you could try the haproxy's leastconn
mechanism.
It will innitially a increase the load on the backend (cache needs to
downloaded 3 times (if not configured to look up the neighbors first)) but
the load on the frontends should always be more or less even.
Does high number of Writing is leading to increase in response time for
delivering the content ?
Of course, it means that more clients are getting the data at the same time
(more disk/network io etc).
rr
_______________________________________________
nginx mailing list
[email protected]
http://mailman.nginx.org/mailman/listinfo/nginx
