What is the biggest disadvantage of Oracle Parallel Server (OPS)?
There are two big disadvantages to Oracle Parallel Server (OPS). Which one is the "biggest" all depends on your perspective. First, OPS is very complex. This complexity does bring high availability and it can bring higher performance. But it is not an easy thing to set up and administer OPS. Second, OPS has a problem known as "disk pinging." To illustrate this, we'll assume that there are two instances in the OPS configuration. Instance1, running on Server1, currently has some dirty data in its buffer cache. An application connected to Instance2 on Server2 wants to access that data. Before it can access that data, Instance1 writes the data back to the shared disk storage. Then Instance2 reads this data into it's buffer cache. This data going back and forth on disk is known as disk pinging and has been known to cause some performance problems.
Oracle addressed this last issue in Oracle9i's Real Application Clusters (RAC), the replacement for OPS with a product called Cache Fusion. Actually, Cache Fusion appeared late in Oracle 8.1.7's OPS. With Cache Fusion, Instance2 could read the buffer cache of Instance1 through a high-speed interconnect between the two servers. RAC with Cache Fusion eliminated the disk pinging that plagued OPS deployments. This only leaves the complexity issue.