Summary Suncluster 3 And NTP

From: Alan Scott (alan.scott@bbc.co.uk)
Date: Thu May 15 2003 - 04:21:56 EDT


Apologies for long delay, each config trial takes a few days and I
decided to let them run while I was on a weeks break, also management
decided that all had to be switched over to GMT from BST at the last
minute.

My thanks go to...

Tim Chipman
Jay Lessert
Patrik Koch
Darren Dunham
Steve Camp
Tray Valenta
Rekhi Manjeet
Martin Hepworth
Stan.Pietkiewicz
Richard Mackerras

 My thanks also to whoever came up with the idea for this list and those
that maintain it, over five years I've been using it, and it hasn't let
me down yet.

 Original question at end.

  The consensus was that the "fudge" option is what was needed on the
cluster node1, setting this to 6 or greater means that clients will tie
to the bbc time server first as this is at a higher stratum level, which
is what I wanted.

The clusters and the apps servers now all sync to the bbc time server,
and should that become unavailable they will all sync to the the first
node in the cluster.
I was also recommended to remove the line

server 127.127.1.0

Because the clients would start to use there undisciplined local clock.

So now the ntp.conf on cluster node1 contains

# The local (undisciplined Solaris) clock is identified as 127.127.1.0.
#
# Some type of primary clock is important; but, it does not have
# to be the local kernel clock.
#
server 127.127.1.0
#
fudge 127.127.1.0 stratum 8
# Setting to stratum 8 clients should tie to bbc time server first
#
# Use BBC NTP server, will fall back to local peer/clock should this
fail
server bbc.timeserver.com
driftfile /var/ntp/ntp.drift
#

On cluster node2
# The local (undisciplined Solaris) clock is identified as 127.127.1.0.
#
# Some type of primary clock is important; but, it does not have
# to be the local kernel clock.
#
server 127.127.1.0
#
# Use BBC NTP server, will fall back to local peer/clock should this
fail
server bbc.timeserver.com
# One of the "peers" should be "preferred" over the others.
#
peer clusternode1-priv prefer
peer clusternode2-priv
#
# Drift file (optional)
#
driftfile /var/ntp/ntp.drift
#

and finally the ntp.conf file for the apps servers
#
server bbc.timeserver.co.uk
#
server node1
#
driftfile /var/ntp/ntp.drift

 If anybody sees further problems with this please mail and I'll
summarise the summary.

The Original question.

We have a cluster of two Sun Fire 880's(Sunc cluster 3 Solaris 8 ).

Also 4 application servers(Sun fire 280's not in cluster Solaris 8 ).

It is essential that all are synchronized to the same time.

I would like that we are able to use our onsite timeserver as primary,

falling back to one of the application servers should the onsite

timeserver become unavailable, or possibly another onsite timeserver.

The cluster ntp.conf file implies that they should always be synced

against each other, I am unable to figure out the required format for a

ntp.conf that would allow me maintain this requirement while having the

onsite timeserver as the "preferred".

Basically I need a ntp.conf for the clusters and the apps servers

that...

uses onsite timeserver as primary, falls back to one of the apps servers

should this fail.

It has to be one of the apps servers, because cluster nodes cannot be

configured as NTP servers.

Is it just a matter of changing the server line to...

server time-server preferred

server backup-timeserver

And adding same to ntp.conf on the apps servers?.

all are running Solaris 8.

Currently the config for the Sunclusters is...

# The local (undisciplined Solaris) clock is identified as 127.127.1.0.

#

# Some type of primary clock is important; but, it does not have

# to be the local kernel clock.

#

server 127.127.1.0

#

# Cluster nodes - private interfaces

#

# All nodes within the cluster must synchronize within the cluster

# as peers. Time synchronization amongst the nodes is more important

# than the accuracy of the agreed upon time.

#

# Peer communications should always occur across the private

interconnect.

# And, so, private hostnames should always be used. Whenever a change

# is made to the private hostname used for accessing a node, a change

# should also be made to the ntp.conf file on each node. When nodes

# are added to or removed from the cluster, this list must also be

# updated on each node.

#

# One of the "peers" should be "preferred" over the others.

#

peer clusternode1-priv prefer

peer clusternode2-priv

#

# Drift file (optional)

#

driftfile /var/ntp/ntp.drift

#

# Stats (optional)

#

filegen peerstats file peerstats type day enable

filegen loopstats file loopstats type day enable

filegen clockstats file clockstats type day enable

#

________________________________________
Alan Scott
Solaris System Administrator (Contract)
News Resources, BBC News
Room 1244, BBC Television Centre, Wood Lane, London, W12 7RJ.
Telephone +44 (0) 20 822 59623
Mobile +44 (0) 7803 163046

BBCi at http://www.bbc.co.uk/

This e-mail (and any attachments) is confidential and may contain
personal views which are not the views of the BBC unless specifically
stated.
If you have received it in error, please delete it from your system, do
not use, copy or disclose the information in any way nor act in
reliance on it and notify the sender immediately. Please note that the
BBC monitors e-mails sent or received. Further communication will
signify your consent to this.
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers



This archive was generated by hypermail 2.1.7 : Wed Apr 09 2008 - 23:26:24 EDT