Disjoint array computer

Abstract

A hierarchical array computer architecture comprised of a master computer connected to a plurality of node computers wherein each node has a memory segment. A high speed connection scheme between the master computer and the nodes allows the master computer or individual nodes conditional access to the node memory segments. The resulting architecture creates an array computer with a large distributed memory in which each memory segment of the distributed memory has an associated computing element; the entire array being housed in a blade server type enclosure. The array computer created with this architecture provides a linear increase of processing speed corresponding to the number of nodes.

Claims

1. A disjoint array computer (DAC), comprising: a DAC master computer including a host port and a disjoint master computer communication channel communicatively connected to a disjoint node computer, wherein the DAC master computer has direct access to the disjoint node computer via the disjoint master computer communication channel; wherein the disjoint node computer includes a node device port connected to the host port via the disjoint master computer communication channel, wherein the node device port supports a plurality device architecture that allows the DAC master computer to view the host port as being connected to a plurality of devices via the disjoint master computer communication channel; wherein the disjoint master computer communication channel provides for: a map to a mass storage of the disjoint node computer, a command channel between the DAC master computer and the disjoint node computer, and a USB multi-gadget communication channel used by the DAC master computer to identify and initialize the disjoint node computer; and wherein the mass storage is accessible either by the DAC master computer or alternatively by the disjoint node computer via commands sent by the DAC master computer over the disjoint master computer communication channel but is inaccessible and physically isolated from other disjoint node computers.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) FIG. 1: Prior Art—A typical blade server enclosure.

(2) FIG. 2: Prior Art—A blade server which provides USB port access to the server blades.

(3) FIG. 3: Prior Art—The Sauermann Parallel APL Machine

(4) FIG. 4: DAC Computer Architecture showing the connection between the master computer and the node computers.

(5) FIG. 5A: the DAC master computer being connected to an array of memory segments.

(6) FIG. 5B: The DAC master computer being connected to an array of computing nodes.

(7) FIG. 6: The preferred embodiment of a two level DAC computer wherein a single USB multi-gadget channel between the level 1 master computer and each level 2 node computer provides both a storage link and communication link between the master computer and its connected node processor.

DETAILED DESCRIPTION

(8) The present invention relates generally to computer architecture and particularly to a hierarchical array computer. The following description is presented to enable one of ordinary skill in the art to make and use the invention and is provided in the context of a patent application and its requirements. Various modifications to the preferred embodiment and the generic principles and features described herein will be readily apparent to those skilled in the art. Thus, the present invention is not intended to be limited to the embodiment shown but is to be accorded the widest scope consistent with the principles and features described herein.

(9) A short acronym, “DAC”, will be used to describe the present invention. The name comes from the phrase, “Disjoint Array Computer”, because a DAC computer is a parallel array computer wherein an array of computing nodes are connected to a master computer using high speed data channels that do not share a common bus. That is, the computing nodes are disjoint. FIG. 4 shows an embodiment of a hierarchical two level DAC computer 400.

(10) In FIG. 4, the high speed data paths 402a-n provide the dual function of (a) a communication channel between the master computer and each of the node computer 403a-n, and (b) providing the master computer with N number of mass storage memory segments 404a-n. Each memory segment 404a-n is only addressable by either the master computer 401 or alternatively addressable by the individual computer nodes 403a-n. By using the communication channels 402a-n, the Master computer and the nodes which level of computing will have sole access the mass storage memory segments 404a-n. The DAC computer 400 of FIG. 4 provides a unique memory mapping structure in that it allows the creation of a large segmented memory for the level one master computer 401 whereby the memory segments are accessed using high speed disjoint data channels 402a-n connected to the computing nodes 403a-n.

(11) FIG. 5A shows how the shared memory appears to the master computer when the master computer has sole access to all of the N number of memory segments. FIG. 5B shows how the master computer views the array of Level 2 nodes when all of the nodes have sole access to the memory segments.

(12) This embodiment provides the means of creating a distributed mass storage architecture by allowing many multiple memory blocks each with its own controller to be viewed as a large distributed memory store controlled by one memory manager. Thus the DAC memory structure provides a memory store with a linear expansion capability using relatively small memory block increases. For example, if the DAC computer of FIG. 4 has 32 nodes each with 1 TB of SSD Flash drives and these are replaced by 4 TB of SSD drives, the total memory as seen by the level one DAC computer has increased from 32 TB to 256 TB.

(13) A major benefit of such a distributed storage architecture with each memory segment having its own controller is that each memory segment can be accessed much faster than if the same amount of storage resided in one large store being accessed by a single memory controller.

(14) A second major benefit is that the computing nodes are closely coupled to the data for which they are tasked to process and that this coupling excludes any conflict with any other node or the master computer.

(15) The dual nature of the individual memory segments to be owned completely by either by the master computer or alternatively by an isolated computing node is an important element of the DAC design. It allows the software programs in both the master and node computers to completely own the memory upon which they are accessing and to not have to deal with interactive problems typically associated with multi-port memory.

(16) The architecture described above and shown in FIG. 4 and FIGS. 5A and 5B is what we refer to as disjoint computing using disjoint memory.

(17) Because the same area of the DAC shared memory of FIG. 4 should not be accessed at the same time by both the master computer 401 and one of the computing nodes 404, means must be provided between the master computer and any given computing node which allows the master computer to coordinate the exclusive access to the shared memory.

(18) FIG. 6 shows a version of the preferred embodiment of FIG. 4 of the DAC computer architecture in which the high speed data channel between the level one computer 600 and a level 2 node 603 is USB serial channel 602. In this embodiment, the computing node USB device port 604 is connected to the master computer USB host port 601. The node USB device port driver supports a three way USB multi-gadget architecture that allows the master computer to view its USB host port 601 as being connected to three different USB gadgets; namely a memory storage block I/O device 604b, an Ethernet port 604a, and a TTY port 604c. In this particular embodiment, the USB block I/O device 604b is mapped to a physical storage drive of the computing node 606. The USB gadget virtual Ethernet port 604a can be used as the command channel between the master computer and the computing node. The USB gadget TTY port 604c can be used by the master computer to identify and initialize the computing node.

(19) Although USB connections were used in the above embodiment to connect the master computer to the computing nodes, other physical buses with supporting software drivers could be used to connect the master computer to the node computers in a similar manner described above including the use of custom FPGA's connected to the respective PCle buses on the master and node computers.

(20) Web blade servers are often used to process high speed web traffic. The blade server architecture that was discussed earlier is similar to the preferred embodiment of the DAC computer shown in FIG. 4. Both implementations are normally designed to fit in a standard server rack. Both use a common power supply to power individual processing nodes. And both have a manager to control the processing nodes.

(21) However, there are major differences between a standard web blade server and the preferred embodiments of the DAC computer which have been presented. The main communication channel of the blade server for data transfers is a high speed bus connected to all the blades including the blade management module. Alternatively, the DAC computer uses isolated data channels between its master computer and the computer nodes. In addition, the memory structure of the DAC computer is a novel storage memory that is used to provide both a large segmented memory for the DAC master computer and alternatively smaller main memory segments for the DAC computing nodes. And finally, the DAC nodes have no direct access to the external memory outside of the DAC computer enclosure whereas the Web Server blades directly address external memory through a network connection such as an Ethernet network.

(22) The hierarchical DAC computer architecture achieves unparalleled simplicity in both hardware and software design by keeping complete orthogonality of the memory and communication between the DAC master computer and the DAC computing nodes.

(23) In addition, the physical placement of the segmented storage memory at the computing nodes allows extremely large storage to be easily controlled by a multitude of memory controllers.

(24) Although the present invention has been described in accordance with the embodiments shown, one of ordinary skill in the art will readily recognize that there could be variations to the embodiments and those variations would be within the spirit and scope of the present invention. Accordingly, many modifications may be made by one of ordinary skill in the art without departing from the spirit and scope of the appended claims.