Guaranteed inter-process communication
View Patent ↗A multiprocessor system includes multiple processors executing multiple processes that send messages to each other. The system uses a connectionless single-window inter-process communication (IPC) protocol to avoid the high message-processing overhead of a connection-based protocol like TCP. The IPC protocol uses a central message-receive service that runs in the kernel on each processor to handle error-checking and sequence number checking and acknowledge received messages, while the single-window mechanism avoids out-of-order message reordering/buffering complexity. Since each process sending a message does not need to wait for the receiving process (which may be otherwise preoccupied or not even the focus of the receiving processor when the message is received and queued) before receiving acknowledgments, the protocol greatly reduces single-window message latency for all processes without resorting to complex connection-based multi-packet transmission windows.
1. A method of operating a first processor, the method comprising:
receiving, at a connectionless message service supporting multiple processes running on the first processor, a first message from a second processor, the first message addressed to a first process running on the first processor, the first message containing header information including a sequence number and error detection coding;
the connectionless message service performing error detection and sequence number checks on the first message, based on the sequence number and error detection coding;
when the first message passes the error detection and sequence number checks on the first message, the connectionless message service attempting to queue the contents of the first message in a first queue assigned to the first process; and
when the first message is successfully queued, the connectionless message service creating and sending, to the second processor, a first acknowledgment message indicating the first message was successfully received.
2. The method of claim 1 , wherein the connectionless message service utilizes the Uniform Datagram Protocol (UDP) for message transport and the Internet Protocol (IP) for delivery to the second processor.
3. The method of claim 2 , the connectionless message service using information existing in a UDP header attached to the first message to perform at least one of the tasks including performing error detection, determining the identity of the first queue, determining the identity of the first process, and creating the first acknowledgment message.
4. The method of claim 2 , the connectionless message service using information existing in an IP header attached to the first message to perform at least one of the tasks including performing error detection, determining the location of a saved sequence number to be used for the sequence number check, and creating the first acknowledgment message.
5. The method of claim 1 , the connectionless message service interpreting an Inter Process Communication (IPC) header attached to the first message to process the message.
6. The method of claim 5 , the connectionless message service removing the IPC header in the first message as queued in the first queue.
7. The method of claim 5 , the IPC header containing a checksum field and a sequence number field, the connectionless message service using the checksum field to perform the error detection check and the connectionless message service using the sequence number field to perform the sequence number check.
8. The method of claim 5 , the IPC header containing a sequence number field and a sequence number reset flag, the connectionless message service performing the sequence number check by:
when the sequence number reset flag is unset, retrieving a stored sequence number value associated with the first process and the second processor, comparing the stored sequence number value to a value stored in the sequence number field, and failing the sequence number check when the comparison evaluates false or passing the sequence number check when the comparison evaluates true; and
when the sequence number reset flag is set, storing the value stored in the sequence number field as the stored sequence number value, and passing the sequence number check.
9. The method of claim 1 , further comprising, when the first message is not successfully queued, the connectionless message service creating and sending, to the second processor, a second acknowledgment message indicating the first message was not successfully received, and a code indicating the reason for failure.
10. The method of claim 1 , further comprising the connectionless message service distinguishing whether the first message is a data message or an acknowledgment message corresponding to a second message sent to the second processor by the first process, the connectionless message service handling the first message, when the first message is an acknowledgment message, by performing an error detection check and queuing the acknowledgment message in a second queue assigned to the first process for acknowledgments.
11. The method of claim 10 , the first process calling a send function of the connectionless message service to send messages to other processors, the send function allowing the first process to have extant at any given time one sent, unacknowledged message to each combination of another processor and port number, the send function attaching a sequence number to each unacknowledged message, the sequence number based on the identity of the other processor and the port number, and any last sequence number used for the other processor and port number, the send function allowing the first process to send another unacknowledged message to a previously used combination of other processor and port number once the prior message sent to the combination has been acknowledged.
12. The method of claim 11 , the send function operating a retry timer and resending a previously sent message upon expiry of the timer without receiving an acknowledgment corresponding to that previously sent message.
13. The method of claim 1 , wherein the connectionless message service for receiving, checking, queuing, and acknowledging messages runs in a kernel executing on the first processor.
14. The method of claim 13 , the connectionless message service allowing processes running on the first processor to register a port and receive a corresponding receive queue assignment, the connectionless message service tracking sequence numbers for each combination of registered port and remote processor sending a message to that registered port.
15. The method of claim 13 , the connection message service allowing processes running on the first processor to register an outgoing port and receive a corresponding send queue assignment, the connectionless message service tracking sequence numbers for each combination of registered outgoing port and remote processor and remote port to which a message is sent from that registered port, sending messages from the outgoing port with a sequence number based on the tracked sequence number for the combination of registered port, and remote processor and remote port to which the messages are addressed, and receiving acknowledgments of the packets sent from a registered port.
16. The method of claim 15 , the connectionless message service resending messages for which an acknowledgment is not received within a given timer interval.
17. The method of claim 15 , the connectionless message service sending a pending message from one of the send queues and addressed to a given combination of remote port and remote processor, when an acknowledgment is received for a previous message sent to that given combination of remote port and remote processor.
18. The method of claim 15 , the connection message service allowing each process registering an outgoing port to check the acknowledgment status for messages placed in the send queue assigned to that outgoing port.
19. An apparatus comprising a computer-readable medium containing computer instructions that, when executed by a first processor, causes the first processor to perform a method comprising:
receiving, at a connectionless message service supporting multiple processes running on the first processor, a first message from a second processor, the first message addressed to a first process running on the first processor, the first message containing header information including a sequence number and error detection coding;
the connectionless message service performing error detection and sequence number checks on the first message, based on the sequence number and error detection coding;
when the first message passes the error detection and sequence number checks on the first message, the connectionless message service attempting to queue the contents of the first message in a first queue assigned to the first process; and
when the first message is successfully queued, the connectionless message service creating and sending, to the second processor, a first acknowledgment message indicating the first message was successfully received.
20. The apparatus of claim 19 , the computer instructions for receiving, checking, queuing, and acknowledging messages running in a kernel executing on the first processor.