Distributed virtual memory management

A data processing network, comprising:

• a hub data processing network;
• a first branch integrated-circuit data processing system;
• and
• a second branch integrated-circuit data processing system;
• the hub data processing network comprising:
• a plurality of request nodes, each request node including a local memory, the local memories of the plurality of request nodes accessible as a distributed virtual memory (DVM);
• a plurality of DVM nodes, where the local memory of a request node is accessible by other request nodes via one or more DVM nodes of the plurality of DVM nodes;
• assignment logic configured to assign one or more request nodes of the plurality of request nodes to each DVM node of the plurality of DVM nodes to form a DVM domain for each DVM node;
• an interconnect system operatively coupling the one or more request nodes in a DVM domain to the DVM node in the DVM domain and operatively coupling between the DVM nodes;
• wherein the first branch integrated circuit data processing system is connected to a first DVM domain of the hub data processing network through a first-branch-circuit chip-to-chip gateway of the first branch integrated circuit and a first hub chip-to-chip gateway of the hub data processing network;
• wherein the second branch integrated circuit data processing system is connected to a second DVM domain of the hub data processing network through a second-branch-circuit chip-to-chip gateway of the second branch integrated circuit and a second hub chip-to-chip gateway of the hub data processing network;
• wherein the first branch integrated circuit data processing system and the second branch integrated circuit data processing system are connected only through the hub data processing network;
• wherein the DVM node of the first DVM domain is configured to treat the first hub chip-to-chip gateway, with respect to DVM messages received from the first hub chip-to-chip gateway requesting a DVM operation, and snoop messages sent to the first hub chip-to-chip gateway by the DVM node, as a request node within the first DVM domain;
• and
• wherein the DVM node of the second DVM domain is configured to treat the second hub chip-to-chip gateway, with respect to DVM messages received from the second hub chip-to-chip gateway requesting a DVM operation, and snoop messages sent to the second hub chip-to-chip gateway by the DVM node, as a request node within the second DVM domain.

A method for managing distributed virtual memory (DVM) in a data processing network having a plurality of request nodes and a plurality of DVM nodes, the distributed virtual memory associated with local memories of the plurality of request nodes, the method comprising:

• assigning a plurality of DVM domains, each DVM domain including a DVM node of the plurality of DVM nodes and one or more request nodes of the plurality of request nodes, the plurality of DVM nodes including a first DVM node in a first-request-node DVM domain;
• sending, by a first request node in the first-request-node DVM domain to the first DVM node, a DVM message to request a DVM operation;
• sending, by the first DVM node in response to the DVM message, a first category of snoop message to one or more peer DVM nodes in DVM domains other than the first-request-node DVM domain;
• sending, by a second DVM node of the one or more peer DVM nodes responsive to receiving the first category of snoop message, a second category of snoop message to request nodes within the domain of the second DVM node and a third category of snoop message to a DVM node of a second integrated circuit, wherein the first chip-to-chip gateway is operatively coupled to the second DVM node and wherein the snoop message is sent via the first chip-to-chip gateway of the first integrated circuit;
• receiving, by the first chip-to-chip gateway a snoop message from the second integrated circuit;
• sending, by the first chip-to-chip gateway in response to the snoop message a DVM message to the second DVM node;
• sending, by the second DVM node responsive to receiving the DVM message, a fourth category of snoop message to one or more peer DVM nodes in DVM domains other than the DVM domain of the second DVM node;
• sending, by a peer DVM node responsive to receiving the fourth category of snoop message, a second category of snoop message to request nodes within the domain of the peer DVM node.