Deterministic distributed multi-cache coherence method and system

Claim 2

A cache accessing method for a multiprocessing computer system having a plurality of subsystems coupled to each other via a global interconnect, each of said subsystems including a main memory, a cache and a global interface and wherein each said caches includes a plurality of cache lines, the method comprising: sending a first request from a first subsystem for a cache line of a second subsystem via said global interconnect; receiving said first request at said second subsystem; blocking all new requests for said cache line of said second subsystem sent via said global interface and received after said receiving said first request; receiving a completion acknowledgment from said first subsystem indicating that said first request has been serviced, said first subsystem sending said completion acknowledgment responsive to receiving data corresponding to said first request and an invalidate acknowledgment from a third subsystem; terminating said blocking responsive to said receiving said completion acknowledgment; queuing said all requests received after said receiving said first request until said first request has been serviced; and processing a second request for said cache line of said second subsystem.

Claim 3

The method of claim 1 wherein said request is a read request.

Claim 4

The method of claim 2 wherein said read request is a read-to-share request.

Claim 5

The method of claim 2 wherein said read request is a read-to-own request.

Claim 6

The method of claim 2 further comprising the step of: sending data from said second subsystem to said first subsystem in response to said read request.

Claim 7

The method of claim 5 further comprising the steps of: sending an invalidate request from said second subsystem to a third said subsystem in response to said read request; and sending an invalidate acknowledgment from said third subsystem to said first subsystem in response to said invalidate request.

Claim 8

The method of claim 2 further comprising the steps of: forwarding said read request from said second subsystem to a third said subsystem; and sending data from said third subsystem to said first subsystem.

Claim 9

The method of claim 7 further comprising the steps of: sending an invalidate request from said second subsystem to a fourth said subsystem in response to said read request; and sending an invalidate acknowledgment from said fourth subsystem to said first subsystem in response to said invalidate request.

Claim 10

The method of claim 8 wherein said blocking step is terminated when said second subsystem receives a completion acknowledgment from said first subsystem indicating that said data and said invalidate acknowledgment have arrived at said first subsystem.

Claim 11

The method of claim 2 wherein said second subsystem includes a directory and said step of receiving said read request includes the step of updating said directory.

Claim 12

The method of claim 1 wherein said request is a write-back request.

Claim 13

The method of claim 11 further comprising the steps of: sending a write-back-permission from said second subsystem to said first subsystem in response to said write-back request.

Claim 14

The method of claim 12 wherein said blocking step is terminated when said second subsystem receiving a completion acknowledgment with data from said first subsystem indicating that said write-back-permission has arrived at said first subsystem.

Claim 15

The method of claim 11 wherein said second subsystem includes a directory and said step of receiving said write-back request includes the step of updating said directory.

Claim 16

The method of claim 1 wherein said second request for said cache line of said second said subsystem includes a read request and a write-back request and said queuing includes a queuing of said read request in a read queue and said write-back request in a write queue.

Claim 17

The method as recited in claim 1 wherein said all new requests include said second request for said cache line of said second subsystem sent via said global interface.

Claim 18

The method as recited in claim 16 wherein said second request for said cache line of said second subsystem sent via said global interface originates from said first subsystem.

Claim 19

The method as recited in claim 1 wherein said second request for said cache line of said second subsystem sent via said global interface originates from a third subsystem.

Claim 20

The method as recited in claim 1 wherein said all new requests include a third request for said cache line of said second subsystem sent via said global interface and originated from said first subsystem.

Claim 21

The method as recited in claim 19 wherein said all new requests include said third request for said cache line of said second subsystem sent via said global interface and originated from a third subsystem.

Claim 22

The method of claim 19 wherein said third request for said cache line of said second subsystem sent via said global interface includes a read request and a write-back request.

Claim 23

A multiprocessing computer system comprising: a first subsystem including a first main memory coupled to a first local interconnect, a first cache coupled to said first local interconnect, a first processor coupled to said first cache, and a first global interface coupled to said first local interconnect; a second subsystem including a second main memory coupled to a second local interconnect, a second cache coupled to said second local interconnect, a second processor coupled to said second cache, and a second global interface coupled to said second local interconnect; a global interconnect coupled to said first global interface of said first subsystem and said second global interface of said second subsystem; wherein said global interconnect provides a communication path between said first subsystem and said second subsystem and wherein said first global interface is configured to pass a first request from said second subsystem for a cache line of said first cache of said first subsystem sent via said global interconnect and to block all new requests for a cache line of said first cache sent via said global interconnect and received after said first request until said first request for said cache line has been serviced, and wherein said second global interface is configured to transmit a completion acknowledgment to said first global interface upon said global interconnect, and wherein said completion acknowledgment indicates that said first request has been serviced, and wherein said first global interface is configured to terminate blocking said all new requests upon receiving said completion acknowledgment, and wherein said all new requests are received from said global interconnect.

Claim 24

A multiprocessing computer system comprising: a first subsystem including a first main memory coupled to a first local interconnect, a first cache coupled to said first local interconnect, a first processor coupled to said first cache, and a first global interface coupled to said first local interconnect; a second subsystem including a second main memory coupled to a second local interconnect, a second cache coupled to said second local interconnect, a second processor coupled to said second cache, and a second global interface coupled to said second local interconnect; a global interconnect coupled to said first global interface of said first subsystem and said second global interface of said second subsystem; wherein said global interconnect provides a communication path between said first subsystem and said second subsystem and wherein said first global interface is configured to pass a first request from said second subsystem for a cache line of said first cache of said first subsystem sent via said global interconnect and to block all new requests for a cache line of said first cache sent via said global interconnect and received after said first request until said first request for said cache line has been serviced, and wherein said second global interface is configured to transmit a completion acknowledgment to said first global interface upon said global interconnect, and wherein said completion acknowledgment indicates that said first request has been serviced, and wherein said first global interface is configured to terminate blocking said all new requests upon receiving said completion acknowledgment, and wherein said all new requests are received from said global interconnect.

Claim 25

The multiprocessing computer system of claim 22 wherein said first global interface further includes a directory configured to store a status of said cache line.

Claim 26

The multiprocessing computer system of claim 22 wherein said first global interface further includes a blocker configured to block a second request for a cache line of said first cache of said first subsystem sent via said global interconnect and received after said first request.

Claim 27

The multiprocessing computer system of claim 24 wherein said blocker is further configured to block access to said directory configured to store the status of said cache line.

Claim 28

The multiprocessing computer system of claim 25 wherein said blocker includes a queue configured to store said second request for said cache line of said first subsystem sent via said global interconnect and received after said first request.

Claim 29

The multiprocessing computer system of claim 26 wherein said queue includes a read queue and a write queue.

Claim 30

The multiprocessing computer system as recited in claim 26 wherein said queue is further configured to store said third request for said cache line of said first subsystem sent via said global interconnect and received after said first request.

Claim 31

The multiprocessing computer system as recited in claim 28 wherein said store of said third request for said cache line of said first subsystem sent via said global interconnect continues until said first request has been serviced.

Claim 32

The multiprocessing computer system of claim 22 wherein said blocker is further configured to block a third request for said cache line sent via said global interconnect and received after said first request.