Method and apparatus for routing interrupts in a clustered multiprocessor system
AssigneeUnisys Corporation
InventorsPenny L. Svenkeson, Robert J. Gulick, Doug E. Morrissey
Life of this patent
Estimating the term…
Abstract
A method and apparatus for increasing the routing bandwidth of interrupts between cluster manager devices in a clustered multiprocessor system is disclosed. This is accomplished by providing special cluster manager devices that can convert “N” serial messages received from a local APIC to “M” parallel messages, wherein M is less than N. The special cluster manager device then transfers the “M” parallel messages to a receiving cluster manager device. The receiving cluster manager device converts the “M” parallel messages into the original “N” serial messages, and sends the “N” serial messages to the appropriate local APIC within the receiving cluster.
Claims17
A multiprocessor data processing system comprising:
- a first processing cluster having a first cluster manager and one or more first processors, wherein each of the first processors has an associated interrupt controller;
- a second processing cluster having a second cluster manager and one or more second processors, wherein each of the second processors has an associated interrupt controller;
- first messaging means for providing communication between the interrupt controllers of the first processing cluster and the first cluster manager;
- second messaging means for providing communication between the first cluster manager and the second cluster manager;
- first converting means for converting N messages provided over said first messaging means into M message(s), wherein M is less than N;
- and
- first transferring means for transferring the M message(s) to the second cluster manager via said second messaging means.
A multiprocessor data processing system comprising:
- a first processing cluster having a first cluster manager and one or more first processors, wherein each of the first processors has an associated interrupt controller;
- a second processing cluster having a second cluster manager and one or more second processors, wherein each of the second processors has an associated interrupt controller;
- a serial interface for providing communication between the interrupt controller of the first processing cluster and the first cluster manager;
- a parallel interface for providing communication between the first cluster manager and the second cluster manager;
- first converting means for converting serial messages received via said serial interface into one or more parallel messages;
- and
- first transferring means for transferring the one or more parallel messages from said first processing cluster to said second processing cluster via said parallel interface.
A multiprocessor data processing system comprising:
- a first processing cluster having a first cluster manager and one or more first processors, wherein each of the first processors has an associated interrupt controller;
- a second processing cluster having a second cluster manager and one or more second processors, wherein each of the second processors has an associated interrupt controller;
- first messaging means for providing communication between the interrupt controller of the first processing cluster and the first cluster manager;
- second messaging means for providing communication between the first cluster manager and the second cluster manager;
- first converting means for converting one or more original messages provided via said first messaging means into one or more converted message, wherein the one or more converted messages have a different format than the one or more original messages;
- and
- first transferring means for transferring the one or more converted messages to the second cluster manager via said second messaging means.
A method for transferring an interrupt from a first processor in a first cluster of a multiprocessor data processing system to a second processor in a second cluster of the multiprocessor data processing system, wherein each cluster including one or more processor with each processor has an interrupt controller associated therewith, each cluster further having a cluster manager, with the interrupt controllers of the corresponding processors communicating with the corresponding cluster manager via a first interface, and the cluster managers communicating with selected other cluster managers via a second interface, the method comprising the steps of:
- providing an interrupt from the first processor to the corresponding interrupt controller;
- providing N serial messages that correspond to the interrupt onto the corresponding first interface;
- receiving the N serial messages at the corresponding cluster manager;
- converting the N serial messages into M parallel messages, wherein M is less than N;
- and
- transferring the M parallel messages to another cluster manager via said second interface.
Dependencies are inferred from the claim text.