Abstract: Techniques used in communicating messages between processes are described using a shared message buffer pool and incoming message lists associated with each process. Associated with each process is message management information. Using the message management information, incoming messages to a process are retrieved, outgoing messages to another process are sent, and message buffers are allocated as needed from a free message buffer list. The free message buffer list is a shared resource from which processes obtain a free message buffer to send a message to another process. Access to the shared free message buffer list and process message lists is synchronized through discipline imposed upon updating and accessing various fields of the data structure without requiring a global locking mechanism for synchronization of shared resources.

Abstract: A method and apparatus for performing error detection in a network is disclosed. An error counter is stored in a common memory location accessible by all nodes. The error counter includes separate partitions associated with each node in the network. Each partition includes two error counter values representing, respectively, sending and transmission error counts, for a corresponding node. No global locking mechanism is required to synchronize access to the commonly accessed error counter. Rather, access to the error counter is provided by having code executing on each node adhering to rules regarding the types of access to the various partitions in the error counter. Each node may read from any partition, but may only write to its own partition.