If you have processes that run under 6.4, you can continue to build classic Models. If however, you find that you need a new Model with lots of dimensions that are not hierarchical, making it difficult to fit under the 32 core limit without artificial index dimensions, you might consider using Spectre.
A new Spectre cBase, in a new project, on new server hardware, is a good way to start exploring the features. For existing projects, continue using Models until you have a logical reason to migrate your data to cBases.
Integ versus Spectre
ETL data processing, traditionally done using the Data Integrator, might be faster if done in Spectre. If a task can be performed using Spectre, most likely it will be faster than using Integrator—Integrator works with strings, while Spectre works with binary forms of the data, which is faster. A few ETL tasks cannot be done using Spectre builds. If you can replace your Integrator scripts with Spectre Build and Dive scripts, they should be noticeably faster.
When are Models Faster?
When a classic model is built, dives are precomputed. The first dive is faster against a model since the data can be loaded from disk with the answer.
When are cBases Faster?
Once a cBase is loaded into memory, Spectre's use of the cache makes for faster subsequent dives. A cBase is always faster when using these functions:
- Group
- Dimcount
- Multitab
- Time series
- Named group
- User-defined dimensions
- Dynamic dimensions
- Dive past the block factor
- Filtered Calcs
NOTE: Just after a cBase builds, before the cache has been populated, a cBase dive might be slow. See Spectre Caching.