Abstract: A virtual synchrony wide area network (100) which has at least a first local area network (LAN) (110) and a second LAN (120). A first router (114) and a second router (116) are connected to the first LAN (110), and a third router (142) and a fourth router (146) are connected to the second LAN (120). A point-to-point link (152) is connected between the first and third routers, (154) between the first and fourth routers, (156) between the second and third routers, and (158) between the second and fourth routers. Each router is provided with computer program code (206) for controlling the flow of messages through the routers and to maintain local total order with minimal latency.

Abstract: A virtual synchrony wide area network (100) which has at least a first local area network (LAN) (110) and a second LAN (120). A first router (114) and a second router (116) are connected to the first LAN (110), and a third router (142) and a fourth router (146) are connected to the second LAN (120). A point-to-point link (152) is connected between the first and third routers, (154) between the first and fourth routers, (156) between the second and third routers, and (158) between the second and fourth routers. Each router is provided with computer program code (206) for controlling the flow of messages through the routers and to maintain local total order with minimal latency.

Abstract: Multiple individual messages (210, 220, 230, 240) generated by a process in response to an event or incoming message are transmitted through a system (100) in a virtual synchrony environment by assembling the multiple individual messages into a single atomic message (200) and delivering the atomic message (200) to all or none of the destination addresses of each of the individual messages. A destination process does not respond to any of the individual messages until the entire atomic message (200) has been received. Individual messages not intended for a particular process may be removed by a computer or process at the destination.

Abstract: A computer (100) and a method (200) are provided for automatically generating source code (132) readily operable in a fault-tolerant system or a distributed system, from source code (130) non-operable in either a fault-tolerant system or a distributed system. The computer (100) preferably includes a programmer's computer system including a CPU (104), storage media (108-118), I/O interface 142, input device (144), output device (146). The computer (100) analyzes the source code (130), and automatically generates source code (132), which is then stored in the storage medium (132). The source code (132) contains additional lines of code for marshaling, unmarshaling, and marking modified data objects.

Abstract: A Totem ring network (100) includes Totem rings (102, 104) which are geographically separated by relatively long distances. Time latencies between the geographically separated Totem rings (102, 104) are minimized by providing each Totem ring with a dedicated router (122, 132) located geographically proximate to a respective Totem ring. Each router (122, 132) is joined together through point-to-point links (106, 108), and each server (124, 126, 128, 132, 134, 136) on a local area network (120, 130) of a respective ring (102, 104) is joined to the respective Totem ring (102, 104) and router (122, 132).

Abstract: Disclosed is a method and apparatus for providing fault tolerance in Totem Networks by use of redundant fabrics. The above is accomplished in one embodiment of the invention by operating devices on the network in such a way that the devices mark the token to indicate when the token has been switched from one fabric to another in response to a timeout. A Ring Master device on the network determines, based on switching of the token by devices on the network whether a fabric or device on a fabric of the network has failed. In addition, fabrics that have failed are monitored to determine when they have become operational. Retransmission of improperly received messages as per token-message-order protocols are also provided for situations in which the token is received before all messages intended for a given device have been properly received.

Abstract: Disclosed is a method and apparatus for causing CPUs comprising portions of a fault tolerant process group to operate in an active-standby mode when synchronizing newly on-line CPUs and reverting to an active-replication mode when synchronization is complete. The above is accomplished in one embodiment of the invention by continuing to operate the primary processor in the active-standby mode and updating the newly online CPUs in accordance with a single pass intelligent update algorithm. When synchronization is complete, a message is transmitted to all CPUs in the group causing a reversion to an active-replication mode for all CPUs whether primary or standby. Any already synchronized CPUs that were in a standby mode, when the group is switched to an active-standby mode, are only updated by check-point message data as data synchronization updating record messages being supplied to a newly online CPU are ignored by these already synchronized standby CPUs.

Abstract: Disclosed is a method and an apparatus that can provide backup CPU data synchronization in a time minimization process. All external messages received, that are to be processed by a primary CPU 12, are also stored in a message list 24 of the backup CPU 14. Check-point message data is intelligently stored in the backup CPU 14 by first deleting related external messages from backup message list 24 and then creating a record if none already exists and filling only those fields referenced in the check-point message. If a record does exist, only the check-point message data fields are altered in that existing record. When a data synchronization record is received by the backup CPU 14, a check is made to see if such a record has already been created by a check-point message. If not, a record is created in the backup CPU database 22 and all the fields are made to correspond with the received data synchronization record message.