In Conclusion

If you are deploying a new application, your important data—the data that is modified often by the end-users—will be stored in a database such as Postgres, MySQL, or Oracle. The cluster nodes will rely on the database server (outside the cluster) to arbitrate write access to the data, and the locking issues I've been discussing in this chapter do not apply to you.^[27] However, if you need to run a legacy application that was originally written for a monolithic Unix server, you can use NFS to hide (from the application programs) the fact that the data is now being shared by multiple nodes inside the cluster. The legacy multiuser applications will acquire locks and write data to stable storage the way they have always done without even knowing that they are running inside of a cluster—thus the cluster, from the inside as well as from the outside, appears to be a single, unified computing resource.

^[27]To build a highly available SQL server, you can use the Heartbeat package and a shared storage device rather than a shared filesystem. The shared storage device (a shared SCSI bus, or SAN storage) is only mounted on one server at a time—the server that owns the SQL "resource"—so you don't need a shared filesystem.