Scale Out Basics

From Blazegraph
Jump to: navigation, search

Working with the scale-out system is a bit different. High throughput depend on using scattered ordered writes and scattered ordered reads.

  • scattered means that you cut latency by using the resources of the cluster in parallel.
  • ordered means that you are operating on chunks of ordered tuples at a time.

If you are trying to do a lot of point tests on the scale-out indices, then performance will be terrible. If you are doing parallel scans on index shards then performance will be fantastic.

The main class to get connected to the federation is com.bigdata.jini.JiniClient. You pass in the name of the main configuration file as installed to $NAS/config as an argument (see the ClusterGuide). Then you do JiniClient#connect() to obtain a JiniFederation instance. The JiniFederation is a com.bigdata.journal.IIndexManager, which is the same interface that is used by the scale-up architecture. You can use that interface to register indices, to lookup indices, etc. The main interface for working with indices is com.bigdata.btree.IIndex. That interface works for both scale-up and scale-out.