http://docs.oracle.com/cd/E23943_01/bi.1111/e10541/prescatadmin.htm
17.1.3.1 Handling Users of the Catalog
The catalog is designed to scale to thousands of concurrent users. To achieve this scaling, the catalog adheres to the following guidelines:
- The average user typically only reads from the catalog and rarely, if ever, writes to it. In Release 11g, each user is constantly and automatically updating his or her Most Recently Used file, but each user's "read" operations still far outweigh the user's "writes" operations. Therefore, the read-to-write ratio is typically at least 100 to 1.
- While a locking mechanism guarantees that only one user can write to an object at a time, it is rare for multiple users to attempt to write simultaneously to the same object. A feature called "lazy locking" allows users to continue reading an object even when another user is updating that object.
- Modern file systems cache "small" files directly inside the directory record, such that reading any information on a directory simultaneously loads all small files directly into the operating system's memory cache. Therefore, it is good practice to keep files in the catalog "small," especially the frequently "read" .atr metadata files. When these metadata files remain small, then all the .atr files in a directory are loaded into memory with one physical hard disk read. Every file that exceeds the "small" threshold adds another physical hard disk read, which can cause a 100% degradation for each large file. In other words, use care when considering storing arbitrary "Properties" in .atr files.
- Reading an object's .atr metadata file using NFS is far slower than reading it directly from a local disk. For this reason, Presentation Services additionally caches all .atr files internally. This cache can become briefly "stale" when another node in the cluster writes data to the file that is newer than the data that is cached by the current node. Therefore, all nodes are refreshed according to the MaxAgeMinutes element in the instanceconfig.xml, whose default for a cluster is 5 minutes. This default setting commonly achieves the best trade-off between the possibility of stale data and the known performance impact. (The default for an environment without clusters is 60 minutes.)
You can modify the MaxAgeMinutes element for your system. Its parent elements are Cache and CatalogAttributes. Before you modify the element, ensure that you are familiar with the information in Section 3.4, "Using a Text Editor to Update Configuration Settings."
