So I have now had a chance to do some reading an experimentation on my own and Ignite seems to be very impressive in terms of the capabilities it provides. Besides for providing the standard ability of an in-memory data grid to cache data across many remote JVMs, the ability to distribute processing to where the data lives, execute functions on individual requests based on operation type, live based event notification when data changes, the ability to read/write through/behind to data store, ect... Ignite in particular has some excellent built in integration with Hadoop. Data grids provide the ability to plug a persistence layer by writing code, Ignite allows configuration based integration with HDFS, no code. This means that it is possible to support OLTP type of access to data in order to support new kinds of applications. It could also help speed up so long running jobs by ensuring that really hot data is read from memory, or perhaps speed things up by using in-memory scratch space instead of local disk (this seemed like it was possible but was not entirely clear). It also seems to also do all of the same things that Tachyon does. The other key integration point is the ability to run Ignite on Yarn. The compute aspect of some of the acceleration features can be managed by the cluster as if it is native. There definitely seems to be some great synergy between Ingnite and Hadoop.
