3.1 Physically set up your system and install AIX
The first step to prepare a system for HA application is to perform the physical setup. This includes all the physical cabling for power, networks, disk devices, etc. We will not go into detail on how to accomplish this as most of this procedure is routine. However, be sure to follow the guidelines below to ensure that the resulting system is capable of serving as a highly available system.
- Keep the shared disks on separate busses
- Cable each disk bus on each node to a separate SCSI adapter, and make sure that the disk drawer that attaches to the SCSI adapter in slot X of node 1 also attaches to the same port on the SCSI adapter in slot X of node 2. This will ensure that hdiskYY on node 1 is the same physical disk as hdiskYY on node 2.
- Route power cables such that the dual power supplies for each node and disk drawer are supplied from a different power distribution strip.
- Use only the ethernet adapters in the PCI slots, do not use the ethernet cards that are integrated into the motherboards. (This is because any ethernet adapter may fail, and replacing the integrated adapter would involve replacing the motherboard).
- Don’t forget to connect your serial (null-modem) cable. Be sure to use it to connect like-named serial ports on both machines. For example, S1 – S1 or S2 – S2 .
- Be as symmetric as possible (make connections exactly the same way on both nodes).
3.2 Mirror rootvg
Mirror rootvg on each machine. Before this step, your rootvg should contain only one disk (we’ll assume hdisk0), and you should have another internal disk available (we’ll assume hdisk1). It is preferable to have each disk on a different internal bus:
% bootlist -m normal hdisk0 hdisk1
Note that the last step reboots the machine. This is necessary to disable the normal quorum checking for rootvg.
3.3 Diagram the Disk Layout
One key aspect to setting up your disks, volume groups, logical volumes, and file systems properly is knowing which disks to use. This is easiest if you draw a diagram to plan out your volume groups. To make sure your drawing is correct, it is important to know what bus the disks are on. Use the lsdev -Cc disk command, and you will get output similar to the following (for SCSI):
hdisk2 Available 20-58-00-0,0 SCSI Disk Drive
hdisk3 Available 20-58-00-1,0 SCSI Disk Drive
hdisk11 Available 20-60-00-0,0 SCSI Disk Drive
hdisk12 Available 20-60-00-1,0 SCSI Disk Drive
Use the lsdev -Cc adapter command, and you will get something like:
scsi3 Available 20-58 SCSI Controller
scsi4 Available 20-59 SCSI Controller
scsi5 Available 20-60 SCSI Controller
scsi6 Available 20-61 SCSI Controller
By comparing the third column in both outputs, it can be determined that hdisk2 and hdisk3 have SCSI IDs 0 and 1 respectively on the bus attached to scsi3 and that hdisk11 and hdisk12 have SCSI IDs 0 and 1 respectively on the bus attached to scsi5. Therefore, you would most likely want to have hdisk2 mirror hdisk11 and hdisk3 mirror hdisk12.
3.4 Plan Shared File Systems
Before creating your diagram, take note that file systems supporting DCE, Encina SFS, application, Sammi, and Java must reside on shared disks. Other site-specific file systems may also need to reside on shared disks in order to ensure that they are available to application regardless of which node is active. The default list of shared file systems to support the above mentioned components are:
Note that this lists default file system mount points only. Determine the appropriate file systems and mount points for your site before continuing.
The sizing of these file systems is very important. To determine sizing for application-related file systems, see 2.10.3 System Memory and Disk Space. To determine sizing for DCE, Java, and Sammi file systems, refer to the appropriate documentation for those applications.
It is imperative that these filesystems are mirrored in such a way that each mirrored copy is stored on a different bus.
3.5 Mirror Shared JFS logs
AIX’s JFS (Journaled File System) also requires a logical volume on which to store its log. Each volume group with at least one JFS will have its own JFS log logical volume. It is important that these be mirrored across busses as well since JFSs can’t operate without their associated JFS logs.
To create your own JFS logs, instead of using the LVs that AIX creates automatically, do the following after creating the volume group but before doing anything else:
1. Create a Logical Volume of type jfslog with a size of one logical partition (1 LP), sequential update policy, and 2 copies of each LP.
2. Format the Logical Volume as a JFS log using the logform command:
Now any JFS that is added to this volume group will use this JFS log.
3.6 Connect Robotic Tape Libraries
Since an HA application system has two nodes which need to be capable of performing all functions of an application core server, it is important that any connections to robotic libraries be identical on both nodes. For example, if /dev/lmcp0 is a control point for a library on one node, then /dev/lmcp0 also needs to be a control point to the same library on the other node. If your library performs security checking by IP address or hostname, use the IP address and interface name of the service adapter when adding your HA system to the library’s security list.