From 8c938d3f7e364a7c191182cb16ed28af35ee3f02 Mon Sep 17 00:00:00 2001 From: Dina Belova Date: Thu, 3 May 2012 15:40:50 +0400 Subject: Mistake with the documentation about cost function's weight corrected. Change-Id: I131d6abc100682ae78c7ac2edfffadc491b9d437 --- doc/source/devref/filter_scheduler.rst | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-) (limited to 'doc/source') diff --git a/doc/source/devref/filter_scheduler.rst b/doc/source/devref/filter_scheduler.rst index 1a7be47cb..285374b70 100644 --- a/doc/source/devref/filter_scheduler.rst +++ b/doc/source/devref/filter_scheduler.rst @@ -225,11 +225,16 @@ The line with this description looks the following way: **function_name_weight**. As for default cost function, it would be: `compute_fill_first_cost_fn_weight`, -and by default it is 1.0. +and by default it is -1.0. :: - --compute_fill_first_cost_fn_weight=1.0 + --compute_fill_first_cost_fn_weight=-1.0 + +Negative function's weight means that the more free RAM Compute Node has, the +better it is. Nova tries to spread instances as much as possible over the +Compute Nodes. Positive weight here would mean that Nova would fill up a single +Compute Node first. Filter Scheduler finds local list of acceptable hosts by repeated filtering and weighing. Each time it chooses a host, it virtually consumes resources on it, -- cgit