Donald K. Burleson Oracle RAC Tips

PolyServe Matrix Server (MxS)

MxS is a general-purpose cluster file system that runs on a Linux or Windows cluster. It is used for Oracle9i RAC and other HA solutions. MxS supports all Oracle data files and redo logs. It further supports advanced Oracle9i functionality including external tables, export/import from text files, and archived redo log compression. MxS also supports a shared Oracle Home, which simplifies installation, configuration, and maintenance.

MxS provides a pseudo device layer that ensures devices are named consistently across the cluster. It provides various tools for file system maintenance, including fsck, file system suspend and resume, and file system growth.

MxS supports context dependent symbolic links that allow administrators to configure node-specific files and/or directories on the CFS.

MxS architecture is completely symmetric for metadata management. A Distributed Lock Manager maintains cache Coherency. The file system is journaled and, due to its symmetric approach, any node in the cluster can perform recovery from the journal in the event of a node failure. There is no master-slave relationship or metadata server.

Storage Control Layer for IO Fencing

The Polyserve Matrix Server includes a storage control layer (SCL), which prevents non-member machines from accessing cluster resources by using the SAN fabric access control mechanism. SCL can use access control mechanisms in the SAN to control which servers are able to use which resources (e.g., I/O fencing, port disabling at the FC switch).

In addition, a small replicated on-disk membership partition is used to allow groups of servers that are otherwise isolated from one another to communicate exclusively through storage. Thus, in a split-brain scenario, the two (or more) fragments of the cluster can discover one another through the membership partition, and in fact can arbitrate through this partition which fragment should continue to have access to the shared storage. This, of course, would only happen if every LAN in the cluster stopped functioning.

Using this out-of-band fencing is a significant benefit in large clustered environments where the alternative fencing approach known as STOMITH (shoot the other machine in the head) is neither sufficiently reliable nor acceptable.