Distributed Architecture Recommandation

[Martin Kersten <ma...@gmail.com>]

Hi everybody,

    I am starting a bigger project for five (to start with) servers under
our control. Each physical server will have 256GB RAM. Two Servers will run
SSDs for persistent reason. The network should only tolerate one node loss
otherwise it should die.

The preferred programming model would be (replicated) objects that allow
listener support to abstract from messages and everything.

Since this distributed memory / state models failed in the everyday
architecture I am still wondering why.

Our requirements will require Hadoop Spark for search and analyzation and
is the reason why we will start with five servers instead of two.

The database this runs is very small (about 10GB of data) so we dont plan
to use a database at all beside for lazy (async) persistence.

Since we have zookeeper with hadoop (if i remember correctly), I ask my
self if there is a open source project (small understandable) providing the
shared object state requirements without a commerical business model and
the size of hazelcast.

I would love to use a solution with < 25k lines which i can totally
understand.

I dont want to write Messages and communicate with messages. I really like
the idea to keep object states in sync and have a single node to be
responsible for changing the objects it owns and the states of objects are
replicated on request.

Underneeth (more primitive layer) this it is really all comming down to
messaging.

So in the end:

Question 1: What does the Project Zoo around Hadoop/Zookeeper offers in
terms of Messaging / eventing?
Question 2: Is there a project providing object/state synchronization?


Thanks,

Martin (Kersten)