You can install
Cluster Edition versions 15.0.3 and later as shared (via an NFS-based or preferably
CFS-based filesystem) by each of the nodes in the cluster, or privately, where each node has
a copy of the binaries installed locally, not shared in any way with any other node. See
your installation guide for installation instructions.
Cluster Edition version 15.0.1 and earlier must be installed on a shared filesystem. For redundancy purposes, Sybase recommends that customers use a cluster-based file system because NFS-based servers raise availability and performance concerns at client sites.
- • For versions of the Cluster Edition 15.0.3 and higher, you can install in “legacy” mode (meaning, the same way as in 15.0.1) or in “private” mode.
- For “legacy” mode installations, make sure that you use a cluster-based filesystem, not an NFS-based filesystem, for $SYBASE. Also, $SYBASE must be mounted in the same location on each of the nodes.
Sybase recommends that you use a private installation to prevent any confusion and make management of the cluster easier. Installing $SYBASE locally also lays the foundation for possible “rolling upgrades” in a future release of ASE cluster edition and allows you to easily shut down a node, perform OS maintenance and then restart that node without affecting any of the other nodes in the cluster.
Following these guidelines will simplify storage management within the cluster:
When the binaries are installed and the UAF agents are running, run
sybcluster to build your cluster. You only need to run this program on one of the nodes because you can manage the cluster with this application from any node in the cluster.
When executing this application, you will need to pass the hostnames of the UAF agents that you are planning to use in your cluster. For example, “sybcluster -F node1, node2, node3”.
Some things to keep in mind when running this application:
- The application doesn’t check to see if the device sizes that you specify are large enough to successfully run the disk init program. If you don’t know the exact size of the devices, you can run dd against them to determine the correct size
- If you make a mistake when entering in any values, make sure at the end to save the information in a file (you will be given the option) and don’t create the cluster yet. Exit out of sybcluster, edit the file that was created by the sybcluster program and then rerun sybcluster. When creating the cluster, you can pass the filename so that you don’t have to re-type everything back in again.
- Linux doesn’t allow you to name the raw devices anything intelligible, so everything is going to be named sequentially (for example, /dev/raw/raw_number). To avoid future confusion, Sybase recommends that you document which device belongs to which device before running sybcluster.
- Create multiple backupsevers to allow for redundancy. The default mode for each backupserver will be “round robin”—good for basic redundancy and high availability. You can change the mode that the backupserver runs in later when the installation completes.
After your cluster is up and running, the following configuration changes from within ASE are recommended:
After your cluster is running, Sybase recommends:
- Have all user databases use 2K logio. This is different than the recommendation for the SMP version; within cluster edition we have shown that the 2K logio size is optimal for a 2K page server (vs the normal doubling of the page size for the logio).
- Turn on the statement cache and enable literal autoparameterization.
- Adjust the CPIC-related configuration parameters (monitor them with MDA) because the default message sizes tend to be too small for larger, OLTP-type of applications.
- Adjust the normal configuration parameters. Keep in mind that they affect each instance of the cluster, so if you have nodes within your cluster that have different resources available to them (memory, CPU’s, etc), you will need to adjust them to fit that specific node.