There are several core issues in distribution. There is no global state space, and the status on each node may not be uniform, and the communication between nodes needs to be across the status space boundary by serialization. Generally only read-only data can do real distribution, because only read-only data can maintain consistency between nodes. Second, information transfer between nodes is unidirectional, there is no lock original language such as Test_and_set (Test_and_set requires simultaneous information bidirectional stream). If you need a distributed lock, there must be a central coordinator, ie there is a bottleneck. Recently popular several concepts involve different aspects of distribution.
. GRID technology mainly handles self-organizational structure of distributed nodes
. Web Service uses XML technology to achieve asynchronous distribution calls, and because of the XML of this meta language, people's intelligence is integrated into the system call level. That is, the remote call may be a program, or it may be to receive user input through a user interface, and then return the user into the web service call result.
. Workflow captures relatively short links between autonomous node behavior. There are "transient" associations between some program steps, which are not stable to abstract Objects in a program, but can create a process processing model (the key to integration is no parameter call, or more accurately said only exists A global variable space is entered from this space and outputs it to this space)
