On 3/27/13 10:21 PM, s madathilthattantav accenture com wrote:
The logic for when a gear is added/removed is encapsulated in the Ruby Daemon (info/bin of the haproxy cartridge). This is unrelated to any of the stats displayed in the stats page.
Take a look at the Ruby Daemon - this is where all the logic for auto-scaling lives. The current algorithm is currently based entirely on the number of concurrent sessions. Memory usage is not part of the auto-scaling or load-balancing algorithm so yes that is expected behavior. On our roadmap we have the ability for users to customize the auto-scaler and to plug and customize other load-balancers (e.g. mod_cluster). For OSE and Origin the auto-scaler can be custoimized by the sysadmin.
The autoscaler is not currently configurable in Online except for the ability for the user to set the floor and ceiling on the number of instances. However it is customizable via the Ruby Daemon for OSE and Origin.
-- Bill DeCoste Principal Software Engineer, Red Hat 978-204-0920 wdecoste redhat com