Memory address mechanism in a distributed memory architecture
First Claim
1. In a distributed processing system comprising read/write path means for carrying memory address and data information and connecting a plurality of nodes, each node comprising at least computer means and local memory connected by a local bus and means for partitioning said memory into one or more partitions, the method of a first node writing in a broadcast mode to one or more other nodes comprising:
- at said first node;
addressing a local memory partition with a local physical address;
converting said partition location physical address to a system physical address;
writing to said system physical address on said read/write path; and
at said others of said plurality of nodes;
detecting said system physical address from said read/write path,converting said system physical address to a local memory partition address, andwriting to the local memory partition at each said other nodes.
2 Assignments
0 Petitions
Accused Products
Abstract
A computer system having plural nodes interconnected by a common broadcast bus. Each node has memory and at least one node has a processor. The system has a dynamically configurable memory which may be located within the system address space of a distributed system architecture including memory within each node having a processor and the memory resident within other nodes. The memory in the system address space is addressable by system physical addresses which are isolated from the physical addresses for memory in each node. The node physical addresses are translatable to and from the system physical addresses by partition maps located in partition tables at each node. Memory located anywhere in the distributed system architecture may be partitioned dynamically and accessed on a local basis by programming the partition tables, stored in partitioning RAMs. The use of the partitioning process permits data to be duplicated throughout a distributed system architecture and permits read cycles for shared data to execute at local bus speeds.
-
Citations
35 Claims
-
1. In a distributed processing system comprising read/write path means for carrying memory address and data information and connecting a plurality of nodes, each node comprising at least computer means and local memory connected by a local bus and means for partitioning said memory into one or more partitions, the method of a first node writing in a broadcast mode to one or more other nodes comprising:
-
at said first node; addressing a local memory partition with a local physical address; converting said partition location physical address to a system physical address; writing to said system physical address on said read/write path; and at said others of said plurality of nodes; detecting said system physical address from said read/write path, converting said system physical address to a local memory partition address, and writing to the local memory partition at each said other nodes.
-
-
2. In a distributed processor system having a plurality of nodes, at least two of said nodes having a computer connected by a primary bus to a local memory comprising a local addressable space having physical addresses, said nodes being connected by an interface to a secondary bus for carrying value and message transactions, said nodes further including an address translator and together defining a system physical address space;
-
the method of establishing distributed common memory comprising; at a first node; establishing in its local physical memory a local memory partition having local memory addresses; identifying a system address space partition, corresponding to said local memory address partition; loading its address translator with said local physical and system physical memory addresses; and informing other nodes to establish similar partitions by message transaction broadcast on said secondary bus; and at each other of said plurality of nodes; responding to said first node information concerning the establishment of a local memory partition by establishing in its local memory a local memory partition; identifying the system physical address space corresponding to said local physical memory address partition; and loading its address translator with said local physical and system physical memory addresses.
-
-
3. In a distributed processor system having a plurality of nodes, at least two of said nodes having a computer connected by a primary bus to a local memory comprising a local addressable space, said nodes being connected by an interface to a secondary bus operative to carry value and message transactions,
the method of establishing shared global memory comprising: -
at a first node at which a distributed common memory is physically resident; establishing in its local memory a local physical memory partition having a local memory address; establishing a system physical address space partition, corresponding to said local memory address partition; informing other nodes that said partition has been established by message transactions broadcast on said secondary bus; and loading an address translator with said local physical and system physical addresses; and at each other of said plurality of nodes at which said distributed common memory is not physically resident, responding to said first node message information by establishing a non-present memory a local memory partition having a local physical memory address; identifying said system address space with said local physical memory address; and loading said address translator with said local physical and system physical memory addresses. - View Dependent Claims (4)
-
-
5. In a distributed processing system having an interconnect bus for carrying value and message transfers among a plurality of nodes in a broadcast mode, each of which comprises a computer means and an address conversion means and at least one of which comprises memory, a method for the computer means at a first node to address the memory in at least one of a plurality of second nodes comprising:
-
generating at said first node a first node logical address; converting said first node logical address to a first node physical address; converting said first node physical address to a system physical address; transmitting said system physical address to said plurality of nodes via said interconnect bus; receiving said system physical address at said plurality of second nodes, including said at least one node having memory; converting said system physical address to a second node physical address; and using said second node physical address to address said second node memory. - View Dependent Claims (6, 7, 8, 9, 10, 11)
-
-
12. A data communications system, having a system address space, comprising:
-
a common communications path for carrying information as message and value transactions; a plurality of communication nodes connected by said path, each of said nodes comprising computing means, memory comprising a local address space identified by local physical addresses, interface means connecting said computing means and memory to said common path, each of said nodes being operative via said interface means to broadcast data and system address information to a plurality of said nodes via said common path and to detect data and system physical address information present on said path, and address translation means operative to convert local memory physical addresses into system physical addresses and to convert system physical addresses to local memory physical addresses, whereby the computing means of a first node can write into memory of a second node as a local first node write operation. - View Dependent Claims (13, 14, 15)
-
-
16. A distributed processing system comprising:
-
read/write path means for carrying information as message and value transactions, a plurality of nodes operative to read from or write to said path means, said plurality of nodes comprising; at least a first node comprising first computing means, first memory means comprising a first local address space identified by first local physical addresses, first local bus means, said first local bus means providing a communications path for said first computing means and first memory means within said node, and first interface means for interfacing said first node to said read/write path means, at least a second node comprising second memory means comprising a second local address space identified by second local physical addresses and second interface means for interfacing said second node to said read/write path means, and means for dynamically partitioning said first memory means and allocating selected ones of the first memory partitions and said second memory means into a system address space identified by system physical addresses, said first computing means being operative to perform read/write operations in said system address space using a local physical address of said node; wherein each of said first nodes comprises address translation means responsive to at least certain of said physical addresses in said first local address space for translating a local physical address into a system physical address in said system address space and address translation means responsive to at least certain of said system physical addresses for translating said system physical addresses to respective local physical addresses. - View Dependent Claims (17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 35)
-
-
33. A distributed processing system comprising:
-
(i) read/write path means for broadcasting memory address and data information, (ii) a plurality of first nodes operative to read from and write to said path means, each first node comprising computing means, local memory and local bus means connecting said computing means, and said local memory and (iii) system memory means, said system memory means being resident on at least one second nodes and comprising at least one of; remote global memory, said memory being resident on said at least one second node and being operatively connected to said communications path and being accessible as system physical address space via said path by the computer means in said first nodes, shared global memory, said memory being resident on one of said first nodes and being accessible as system physical address space by the computer means in any other of said first nodes, and distributed common memory, said memory being resident on one or more selected said first nodes and being accessible as system physical address space having the same physical address for all of selected first nodes. - View Dependent Claims (34)
-
Specification