Académique Documents
Professionnel Documents
Culture Documents
Disk Configuration.......................................................................................................... 1
Introduction
This paper is divided into two parts. Part One provides you the hardware configuration,
Part Two provide you the step-by-step instructions for installing Microsoft Cluster Server
(MCS). Part 3 gives you the step-by-step instructions for installing Oracle Fail Safe and
configuring a database. Here is an overview of the steps required to install MSCS:
1. Hardware Configuration and Set-Up
?? Confirm Hardware is Certified for MSCS
?? Configure Shared Disks
?? Select Disk to be Quorum Disk
?? Configure Network Cards
?? Obtain IP Address and Network Name for Cluster Group and Register in
DNS or HOSTS file
2. Install Cluster Server on First Node and on Second/Additional Nodes
3. Install Oracle Fail Safe
Certified Hardware
Oracle does not specifically certify hardware for Oracle Fail Safe. Instead, you must
ensure that the hardware is on the Microsoft Cluster Server Hardware Compatibility List
(HCL) that is available from Microsoft? . You will find the HCL at:
http://www.microsoft.com/hcl/
Disk Configuration
Disks need only be configured from one node. Do not attempt to write to the disks from
multiple nodes until the clustering software has been installed. Avoid creating software
volumes—any striping or RAID configuration should be done at the hardware level, prior
to configuring the disks in the Disk Management console; this will give you better
performance. Choose a node from which to configure the disks, and open the Disk
Management Console
Partitioning a single physical disk into multiple partitions can be done, but MSCS sees
the entire Physical Disk as a single resource, so the entire disk must always move
together, no matter how many partitions are on it. Therefore, it normally makes sense to
simply create one partition on each Physical Disk. Format all of the shared drives as
NTFS volumes and assign the drive letters as appropriate. Note, in the example below
that we have labeled volumes as either Shared or Private.
1
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
Quorum Disk
MSCS requires that one of your shared disks be assigned as the quorum disk. The
quorum disk assists in handling certain clustering functions. The quorum disk is critical
to resolving ownership of resources should the interconnect go down. Additionally, it
provides an area of physical storage that all nodes can access. The quorum disk does not
require much space, so you should choose the smallest drive possible. Microsoft
recommends a minimum drive size of 500MB. Keep in mind that if the quorum disk fails,
the cluster fails, therefore you may want the quorum disk to be a RAID volume of some
type. It is possible in some versions, to place Oracle datafiles on the same drive as the
quorum disk, Oracle and Microsoft recommend that the quorum disk be kept separate
from any other resource disks.
Decide which shared disk you want to be the quorum disk.
2
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
resources on a regular basis. Not only can this polling result in a large amount of traffic,
but a network glitch could be incorrectly interpreted as a resource failure which could
result in a restart or failover of a healthy resource. Thus, it is better to have a dedicated
network for the resource polling.
Binding Order
With a network card dedicated to the interconnect, and a second card dedicated to the
public network, it is important to ensure that the bindings are set up correctly. Any public
cards which will be communicating with client machines should always be bound first,
leaving the network card for the interconnect bound last of all. This is critical in ensuring
the name resolution works correctly, particularly when nodes are communicating with
each other. If the binding order is incorrect, you may see that a ping of the public host
name resolves to the private IP address. Thus, if a listener is configured to listen on a
host name, it may incorrectly resolve that host name to the private IP address which
means incoming connections from clients will fail.
3
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
4
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
to assign a network name to the cards on the private interconnect. Since these cards
usually are not going to be connected to a DNS server, you should add entries into the
hosts file. You can find the hosts file in the \WINNT\System32\drivers\etc directory. A
popular naming convention is to append ".SAN" to the end of the actual node name, and
use that as the host name assigned to the private card. This convention indicates clearly
that this hostname is on its own subnet, using the private interconnect. If you have two
nodes called RMNTOFS1 and RMNTOFS2, your host file entries might look like so:
Double-check the setup by pinging the public and private names of all nodes in the
cluster, ping each node from itself. Verify that a ping of the public name always returns
the public IP address, and a ping of the private name returns the private IP address:
C:\>ping rmntofs1
Pinging rmntofs1.US.ORACLE.COM [192.1.1.1] with 32 bytes of data:
You will be prompted for the Windows 2000 Advanced Server CD, and then
the Cluster Configuration Wizard will be started.
5
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
6
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
Defining Networks
10. After selecting the quorum disk, you will be presented with a screen on which
you will define the networks. You can name them whatever you choose—
generally, we recommend that you keep it simple and call them "Public" and
"Private". For the "Private" network, you want to ensure that you select the
radio button to enable the network for Internal Cluster Communications
Only. For the public network, you should probably select All
Communications, to provide a certain amount of redundancy.
11. On the next screen, you will determine which network should be used first for
cluster communications, assuming that both networks are functioning.
Be sure that the Private network is first, so that as long as it is functional, the
public network will be configured only as a fallback. It is also fairly common
for some sites to have three or four network cards in each node, so that a
second private network can be defined for the interconnect, again providing
additional redundancy.
If you have more than two cards in each node, configure the networks
according to which order you want cluster communications to fall back in the
event of a failure.
12. For the final step, you will be prompted to enter the IP address that you have
reserved for the virtual Cluster Group. As previously mentioned, this IP
7
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
address is the same that was registered in DNS or the HOSTS file with the
cluster’s Network name that was specified at the outset of the MSCS install.
Type in the IP and ensure that the correct network is chosen. In our example,
the cluster name is RMNTCLUSER, and the IP Address is 138.1.144.117. On
the final screen, be sure to click Finish to complete the cluster installation.
8
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
9
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
Disk Groups
In addition to the Cluster Group, you see in the Figure that you will have a Disk Group
for each additional shared disk besides the quorum disk. These Disk Groups are simply
placeholders for the disk resources—they are not true virtual groups, as they do not have
network names and IP addresses associated. However, ownership of the disk groups can
still be transferred back and forth between the nodes. When a database with files residing
on one of these disks is added to a new group, the disk resource associated will be
removed from the temporary disk group and placed into the database group. At this time,
you will be able to delete the disk group, if you so desire.
10
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
11
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
DCOM Security
In addition to configuring the service logon, the security setup will configure DCOM
access by calling the configuration tool and adding the local SYSTEM account to the
default access permissions list for Distributed COM security. You can view this by
running dcomcnfg at a command prompt and choosing Default Security and editing
Default Access Permissions. In earlier releases of Oracle Fail Safe, the default access
permissions were left untouched. This is normally empty, and thus the SYSTEM and
INTERACTIVE accounts are assumed to have privileges. However, some third-party
applications may add user accounts to the default access list, nullifying any default
permissions. If default permissions are modified, you may experience a hang when
running the Verify Cluster tool unless SYSTEM is explicitly added to the default access
permissions, so starting with the 3.2 release, the Oracle Services for MSCS Security
Setup has been modified to always add the SYSTEM account. See MetaLink Document
ID 155317.1 for more details on this problem.
12
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
cluster key at HKLM\Cluster\Oracle. Last, once the Oracle Database and Oracle TNS
Listener resource types are registered, you will be able to view this under
HKLM\Cluster\Resource Types. If you ever need to remove Fail Safe from a cluster, you
should uninstall it if possible, so that the resource types are unregistered and removed
from the Registry. Uninstalling Cluster Server will remove the HKLM\Cluster key,
forcing you to reregister the Fail Safe resource types after you reinstall Fail Safe. This
can be accomplished by rerunning Verify Cluster, discussed in the next section.
13
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
Fail Safe Manager can be installed on a client machine to allow you remote management
access to the cluster. Previous releases of Oracle Fail Safe required that the Fail Safe
Manager client be the same version as the Fail Safe Server running on the cluster.
However, beginning with the 3.2 release of OFS, the Fail Safe Manager can be used to
manage clusters running Fail Safe version 3.1.1 or later. Thus, in an environment with
multiple clusters, you do not have to upgrade all at once, nor do you need to sacrifice the
manageability of using Fail Safe Manager to manage multiple clusters. Simply ensure
that you have the latest version of Fail Safe Manager on your desktop, and it will work
with the 3.1.x clusters and 3.2.x clusters.
14
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
15
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
16
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
Once you identify your database, right-click it and choose Verify Standalone Database.
You will be prompted for the instance name, parameter file location, and whether you
want to connect using OS Authentication or you want to provide a password. If you
choose OS authentication, Fail Safe will create a local OS group called
ORA_<sidname>_DBA and add the accounts that were specified for the Cluster Service
and the OracleMSCSServices. This allows members to connect only to this particular
instance—Fail Safe will not automatically create the more generic ORA_DBA group, but
it will work if you manually add the accounts to this group instead of a group specific to
your SID.
Creating a Group
We reiterate here that you cannot add the database into the Cluster Group—you must
create a separate group for the database, and you must have a host name and IP address
combination ready. Even though you can use MS Cluster Administrator to create the
group, we recommend that you create it through Fail Safe Manager, as it provides an
interface to add a hostname and IP address into the group. In Fail Safe Manager, right-
click the Groups folder and choose Create. You will be prompted for a name for the
group—this can be any name that you decide on; it need not match the hostname. Type in
the name and an optional description and choose Next:
17
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
18
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
A Failback policy does not have any meaning if there is not a preferred node, because the
Failback is triggered when the preferred node rejoins the cluster. Accordingly, if you
chose to Failback Immediately, this Failback event will be triggered as soon as the
preferred node comes back online. Choosing Prevent Failback on Page 2 implies that
there is no preferred node, so you will not see Page 3 of the Create Group Wizard, which
is where the preferred node for the group is selected.
19
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
newly created group—the wizard configures the group with that address, and then MSCS
is responsible for registering that address with the gateway and directing all network
communications to the appropriate owning node. This virtual address then becomes the
means by which your clients connect to the virtual server and communicate with the rest
of the resources that will ultimately be added to this group. As such, this network name
and IP address combination must be unique on your network, even among other virtual
address that already exist, and it must resolve successfully and be accessible by any
clients that wish to access the database.
Choose Yes in answer to the Add Virtual Address question, and the Add Resource
Wizard will be initialized. You will be prompted to select which network you want to add
the virtual address from. In most cases, you will be choosing the public network, which
allows your clients to access the network. Theoretically, though, if the only client is an
application tier, which runs on one of the other cluster nodes, you could select the private
cluster network.
The network name and address that you supply must be valid on one of the subnets tied
to a physical card. As an aside, it is possible to have multiple IP address and network
name combinations existing in a single group, and it is also possible to have these IPs be
on different subnets, to provide further redundancy and load balancing. However, a
virtual IP address must always be on the same subnet as at least one physical card within
the cluster. Thus, having two IP addresses in a group that are on different subnets would
20
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
require two different physical network cards, each with an IP address on the respective
subnets used by the virtual IP.
21
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
Choose the appropriate network for the initial virtual IP, and then put in the host name
that you have predefined in DNS or your hosts file. If this is set up correctly, the IP
address should be filled in automatically. If not, you will get an error indicating that the
host name does not resolve to an IP address. Another common error here is to put in the
existing host name of the Cluster Group. If you do so, this will fail with an FS-11221
error, indicating that this network name is already in use. Duplicate network names, of
course, are not allowed. The group will still be created, but it will not have a virtual
address assigned. You must then go back to Fail Safe Manager, right-click the empty
group, and choose Add Resource to Group.... The Add Resource Wizard will be initiated
again, and you can choose Virtual Address from the list of available Resource Types, this
time selecting a new network name and IP address combination not currently in use
anywhere on your network.
22
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
Once you have verified this information, continue on to the next screen. Here, you will
define the network service name, the instance name, the database name (as defined by
DB_NAME in the init file), and the location of the parameter file that you wish to use.
The next page is the Database Authentication page. If you previously ran the Verify
Standalone Database procedure and specified that you wanted to use OS authentication at
that time, then it is assumed that you are doing so again when the database is actually
added to the group. If you have not run Verify Standalone previously, or if you chose to
use the SYS account for authentication, then you will be asked again. (Internal is still
offered as an option for backward compatibility, because the 3.2 release of Fail Safe
Manager will support Oracle8i and Oracle 8.0 databases.) If you choose OS
authentication here, again, an OS group called ORA_<sidname>_DBA will be created,
and the logon accounts for both the Cluster Service and the OracleMSCSServices will be
added to this group. If you had done this during the Verify Standalone operation this
group will already exist.
Next, you will still be asked if you want to maintain a password file on all nodes of the
cluster. This is recommended if you want to allow access via the password file, but you
do not want to add certain OS users to the ORA_DBA group. (Refer to Chapter 4 for
more information on using a password file.) The key thing to realize here is that if you do
not use OS authentication, then you must ensure that any changes to the password file are
propagated to all nodes in the cluster. The polling that is done by the Cluster Service uses
23
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
this information to connect, and if the password is wrong on one of the nodes, the polling
may fail, or the database may not be able to come online at all.
24
Microsoft Cluster Server and Oracle Fail Safe Quick Start Guide
25