OpenStack Compute (nova)

XenServer bandwidth usage is inaccurate.

Bug #1055737 reported by Monsyne Dragon on 2012-09-24

This bug affects 1 person

Affects		Status	Importance	Assigned to	Milestone
	OpenStack Compute (nova)	Fix Released	Undecided	Monsyne Dragon	OpenStack Compute (nova) 2013.1 "grizzly"

Bug Description

The bandwidth information reported by nova for instances running on xenserver does not seem to be reliable.
After much testing, the data is accurate for some instances, and off for others, with no discernible pattern. It seems to be due to the fact that the underlying data source (the statistics RRD files provided by XenServer) are not reliable, (and are fairly poorly documented).

Monsyne Dragon (mdragon) on 2012-09-24

Changed in nova:
assignee:	nobody → Monsyne Dragon (mdragon)

Revision history for this message

OpenStack Infra (hudson-openstack) wrote on 2012-09-24: Fix proposed to nova (master)

Fix proposed to branch: master
Review: https://review.openstack.org/13594

Changed in nova:
status:	New → In Progress

Revision history for this message

OpenStack Infra (hudson-openstack) wrote on 2012-09-29: Fix merged to nova (master)

Reviewed: https://review.openstack.org/13594
Committed: http://github.com/openstack/nova/commit/8887f10c66bca248f289db8f834ae8f36f9a03a1
Submitter: Jenkins
Branch: master

commit 8887f10c66bca248f289db8f834ae8f36f9a03a1
Author: Monsyne Dragon <email address hidden>
Date: Mon Sep 24 19:38:09 2012 +0000

Collect more accurate bandwidth data for XenServer

    This changes the method used to poll xenserver for bandwidth data.
    The reccomended way of collecting such data from xenserver (namely the
    RRD files provided by the hosts) do not seem to be reliable, they
    will sometimes be correct, often will be signifigantly under (> 10%),
    and occasionally will show artifacts, such as phantom 4gb bandwidth
    'spikes'.

    This patch changes that to use the much simpler method of simply polling the
    byte counters on the VIF network devices on the host. (We have old non-nova
    code that does that on xenserver, and that method is known to work).

    This should also make it much easier for other hypervisors other than
    xenserver to implement bandwidth polling, as polling the counters is a rather
    more universal method.

Fixes bug 1055737

Change-Id: I6a280d8bbfcc74914f888b11bc09349a270a5f58