Memory Cache Size | Vantage BYOM - Memory Cache Size - Teradata Vantage

Vantage BYOM caches models to eliminate the need to load a model for every query that uses the model, which improves performance. When the cache needs room for a new model, it removes the least recently used model or models. If all cached models are being used, the new model is loaded into the memory (if there is room in JVM memory) but not cached.

When the parameter CacheMaxLimit has its default values, the cache can hold a model of at most 1.2 GB. This model size corresponds to a model file size (BLOB size in a database table) between 436 MB and 1.2 GB, depending on the model type. Different model types have different expansion factors in JVM memory. The largest expansion factor is 2.75 (corresponding to 1200/2.75 = 436 MB), but Dataiku and DataRobot models can be larger. For example, DataRobot models can be up to 10x (such as, in random forest models).

If the cache is too small for the model or models you want to cache, you can enlarge the cache by increasing the value of CacheMaxLimit. Enlarging the cache may require increasing the size of JVM memory to fit the cache (see JVM Memory Size and JVMMemStats Function). Teradata recommends that the cache uses at most 60% of JVM memory.

To change the cache configuration parameter, use the CufConfig JVMOptions field.

The following example increases the cache size limit to 3 GB.

JVMOptions: -DcacheMaxLimit=3g