GridGain

GridGain uses a shared nothing architecture with data nodes that hold both data and indexes and client nodes that act as connection points. The former process incoming requests along with computations. Partitioning is supported with automatic cluster rebalancing whenever a node is added or removed. Support is provided for ODBC, JDBC and REST and a wide variety of languages are supported, including ANSI standard SQL. Transaction processing comes with ACID guarantees while strong consistency is provided along with support for two-phase commit.

Fig 02 Illustrating the Continuous Learning Framework

In so far as Figure 1 is concerned, GridGain enables the ingestion of streaming data through a Kafka connector, and there is also support for both Spark DataFrames and Spark Resilient Distributed Datasets (RDDs). Perhaps most interesting, however, is the Continuous Learning Framework (CLF). This is illustrated in Figure 2. The point is that machine learning and other analytic models can be trained and tested in the platform, which means that those models can be updated directly rather than periodically, which tends to be the case when models have to be trained and tested in an outside environment. This also lends itself to environments where you want to embed analytics “in process” within an otherwise transactional or operational application. While it will often be the case that you want to run these analytics within the GridGain environment, this does not have to be so. You can, for example export models if, for example, you wanted to run these on GPUs for extra processing performance.

While we have already mentioned support for Spark (and Spark ML), GridGain is also integrated with TensorFlow for deep learning. While the company does not specifically support PMML (predictive modelling mark-up language) it can leverage the PMML support provided by TensorFlow or Spark though it would not help if you wanted to import a model derived using SAS, as an example.

The big benefit of using a data grid is that you can simply add it to your existing environment without any of the pain of replacing your current database. On the other hand, there are times when you positively want to replace existing systems and, of course, there are greenfield developments where you need a new database from the get-go. The optional deployment of GridGain as an in-memory database, as opposed to a pure data grid, offers the flexibility to support all of these scenarios.

On top of that, we are impressed with the Continuous Learning Framework. In hybrid environments where you want to support concurrent transactional and analytic processing, there are two generic use cases: analytics (typically dashboards) that run in parallel to operational functions and analytics that are embedded (in-process) in operational applications. Everybody in this space supports the former but GridGain is one of relatively few vendors that offer specific capabilities to support the latter.

The Bottom Line

There is lots of competition in the market for in-memory databases but that is much less true for in-memory data grids. In the latter case, Apache Ignite is by far the most widely deployed offering and GridGain is thus positioned as the market leader in this category: if you want to augment your existing environment with in-memory technology GridGain is an obvious possible choice.