History[06]

Ignite was originally the proprietary GridGain DBMS. It was then open-sourced in late 2014 and then accepted in the Apache Incubator program.

Checkpoints[07]

According to Ignite's confluence page, consistent checkpointing has been implemented in the system. The fuzzy checkpointing may be implemented in future releases.

Compression[08]

There is no data compression implemented in ignite now. However, ignite suggests the Index Prefix Compression in its future enhancement plan.

Concurrency Control[09]

According to Ignite's documentation v2.7.6, MVCC now is released as beta versions in Ignite version 2.7. In MVCC, Ignite will create a logical snapshot of data for each transaction. The snapshot is consistent throughout the transaction and the transaction can only view and modify data in this snapshot. For example, if an entry is read by transaction A and updated by transaction B when transaction A tries to update the entry again, Ignite will throw an exception and transaction A will need to be retried.

Data Model[10]

Ignite uses a key-value store. And Ignite determines the location of the data using a pluggable hashing algorithm. It means every client can determine where the value is stored by plugging the key into a hashing function.

Foreign Keys[11]

Ignite does not support foreign key constraints.

Indexes[12]

Ignite uses B+tree data structures to store regular indexes. Ignite supports primary and secondary indexes, but it will only enforce the uniqueness of primary indexes.

Isolation Levels[13]

Ignite supports the following three isolation levels: READ_COMMITTED, REPEATABLE_READ, and SERIALIZABLE.

Joins

Logging[14]

Ignite uses a write-ahead log which logs logical database modifications and page changes.

Parallel Execution[15]

It supports distributed collocated processing, where the operation can be distributed to each node.

Query Interface

Storage Architecture[05]

Apache Ignite is an in-memory (or memory-centric) distributed database. But there is a native persistence option in Ignite. If you turned it off, it is a pure in-memory store. When the option is turned on, it stores both data and indexes on disk. In this mode, a subset of data is stored in memory. Additionally, one benefit of this mode is that the indexes don't need to be rebuild when the system restarts since they are persisted on the disk.

Storage Model

Storage Organization[16]

Ignite used a page-based memory architecture, where memory is split into pages of fixed size. The pages are stored in RAM and are organized in a special hierarchy on disk.

Stored Procedures[17]

Ignite support access to the contents of the result generated by batches and stored procedures.

System Architecture[18][19][06][15]

Ignite uses a shared-nothing, node-based cluster architecture. The nodes can automatically discover each other. It supports third-party persistence and native persistence. For third-party persistence, it acts as a in-memory cache on top of a third-party database. For its native persistence, it acts as a in-memory (memory-centric) distributed database. It also supports collocated processing, where the computation on data is done on the node where the data is stored.

Views[20]

Ignite has some built-in views that contain information about cluster nodes.