blob: f974452fdeced46e9a82782bdc487c6f12d75ba9 [file] [log] [blame]
What is Hyracks?
Hyracks is a partitioned-parallel platform for running data-intensive computation on a shared-nothing cluster of commodity machines.
Hyracks Concepts
Hyracks employs a client-server architecture. On the server side, the software module that is responsible for interacting with clients, keeping track of and dispatching work
to other machines in the cluster is called the Hyracks Cluster Controller (CC). There is one CC per logical Hyracks cluster. The module that executes on the worker machine
and interacts with the CC to receive work and act on it, is called the Hyracks Node Controller (NC). Every NC in a single Hyracks cluster has a unique logical name. When an
NC is started, it is provided the address of the CC whose cluster it must join. Although it is sufficient to run one instance of the NC on a physical machine, it is possible
to run multiple instances of NCs (ofcourse each NC has a different logical name) on the same physical machine -- often used for simulating a cluster on a single machine
to facilitate testing.
Hyracks clients interact solely with the CC when submitting their jobs. A Hyracks Job is the unit of work that a client can execute on the Hyracks cluster. A job is expressed
as a directed acyclic graph (DAG) of Operators connected to each other by means of Connectors. A more detailed description of jobs, operators, and connectors follows in
chapter "Hyracks Jobs".