The MIT J-machine [17, ] combines a general
purpose CPU with a network controller. (Similar to the nCUBE
processors [67, ].) Upon arrival of a message on a
J-machine node, the CPU dispatches a handler to process the message.
The handlers are fine-grained threads. Dispatching is done by
the hardware and, therefore, extremely fast (less than 
).
In contrast to the Stanford FLASH project, there is only one CPU to handle message traffic and user applications. On the other hand, user applications and message processing are tightly integrated, and applications provide the threads to handle incoming messages. From an applications point of view, this makes the J-machine more flexible, since the protocol processor code in the MAGIC cannot be changed by an ordinary user.