Product DocsMenu

Coveo Platform 7.0 >
Administrator Help > Coveo Enterprise Search 7.0 > CES Features versus Hardware Resources

Relation Between CES Features and Hardware Resources

CES involves several processes that are running concurrently on one or more servers. Depending on the size of your Coveo installation, the activated features, and the phases of operation, these processes consume various levels of hardware resources.

The following list describes Coveo processes, components, or features that affect the hardware resources.

  • Document conversion is entirely done in parallel. The greater number of CPU cores the better.

  • Querying requires several CPU cores to perform as many steps in parallel.

  • Queries with numerous terms, exact match operators, NEAR operator, or wildcard characters can take significant amount of CPU resources.

Note: You can configure the relative priority of the main, indexing, and crawling processes as well as specify the number of query threads (see Modifying Advanced CES Parameters).

Physical memory (RAM)
  • Indexing uses a lot of physical memory to pre-compute mappings from terms to identifiers. More memory is better.

  • Querying requires a good amount of physical memory for caches.

  • Document conversion typically loads documents in physical memory.

  • Numerous numerical fields often require to be kept in physical memory to achieve good query performance.

  • Facet fields require an amount of physical memory directly proportional to the number of facet values to be cached. Not having enough memory to cache facets is not an option, as query performance would degrade significantly.

  • String or numerical sorting fields have to be set up to be loaded in physical memory. For string sorting fields, the number of field items (cardinality) is what dictates how much memory is needed, the higher it is, the more memory it takes. For numerical sorting fields, cardinality doesn’t matter, only the number of documents in the index does.

Hard disk

  • Indexing is disk I/O intensive. Upgrading the disk subsystem has the most impact for better performance.

  • Querying is a process requiring a fast disk subsystem.

  • Adding many string fields affects the disk subsystem, because it adds a lot of new terms to the index.

  • Facet fields are easier to index when the number of different facet values is low (cardinality). The higher the cardinality, the higher the stress on the disk subsystem.

  • String sorting fields put more stress on the disk subsystem than numerical fields.

  • Document summarization produces a concept list and summary sentences that are added to the index (see Modifying Advanced Converter Parameters).

  • Document conversion accesses hard disk only for very large documents.

People who viewed this topic also viewed