Mesh network topology based on midplane board

Patent metadata

Field	Value
Publication number	US-10694635-B1
Application number	US-201816103653-A
Country	US
Kind code	B1
Filing date	Aug 14, 2018
Priority date	Aug 14, 2018
Publication date	Jun 23, 2020
Grant date	Jun 23, 2020

Abstract

A system can include a midplane board and multiple sleds. Each sled can include a processor and an interconnect part. The interconnect part can be positioned at a lateral edge of the sled. Both sides of the midplane board can include interconnect portions that engage the interconnect parts from the sleds. The interconnect portions of the midplane board can be connected to enable inter-processor communication among processors on the sleds. For example, the midplane board may connect processors from first and second sleds received on one side of the midplane board with processors from third and fourth sleds received on an opposite side of the midplane board. The sleds may connect to power at rear, for example, being translated in a longitudinal direction to engage power and translated toward the midplane board and non-parallel to the longitudinal direction to engage the interconnect parts and portions.

First claim

What is claimed is:

1. A system, comprising:
a chassis sized to fit in a server rack, the chassis defining a space bounded by a front, a rear, a top, a bottom, a left side, and a right side, the chassis further defining a vertical direction between the top and the bottom, a longitudinal direction between the front and the rear, and a lateral direction between the left side and the right side;
a power harness extending in the vertical direction and comprising a first power connector, a second power connector, a third power connector, and a fourth power connector, wherein at least the first power connector is configured to receive in the longitudinal direction to establish connection for transmission of power and to permit translation therein substantially perpendicular to the longitudinal direction;
a midplane board extending in the longitudinal direction and comprising a first interconnect slot and a second interconnect slot on one side of the midplane board and a third interconnect slot and a fourth interconnect slot on an opposite side of the midplane board; and
a set of sleds each respectively comprising a rear edge, a lateral edge, a processor, and a connector positioned at the lateral edge and coupled with the processor, the set of sleds comprising:
a first sled having a first connector received in the first interconnect slot of the midplane board and coupled at a first rear edge to the first power connector of the power harness;
a second sled having a second connector received in the second interconnect slot of the midplane board and coupled at a second rear edge to the second power connector of the power harness;
a third sled having a third connector received in the third interconnect slot of the midplane board and coupled at a third rear edge to the third power connector of the power harness; and
a fourth sled having a fourth connector received in the fourth interconnect slot of the midplane board and coupled at a fourth rear edge to the fourth power connector of the power harness;

wherein the first, second, third, and fourth interconnect slots of the midplane board are electrically coupled to one another to form a mesh network among the processors of the set of sleds.

2. The system of claim 1 , wherein each sled of the set of sleds further comprises a topology link connecting the processor to the connector, wherein the topology link is less than six inches in length.

3. The system of claim 1 , wherein each connector comprises an Ultra Path Interconnect (UPI) interface or a Peripheral Component Interconnect Express (PCIE) interface.

4. The system of claim 1 , wherein each sled comprises two connectors configured to engage the midplane board, wherein a first of the two connectors is coupled with the sled's processor, and wherein a second of the two connectors is coupled with a component other than the sled's processor.

5. The system of claim 4 , wherein one sled's processor is connected to another sled's component, other than the other sled's processor, via a connection provided by the midplane board.

6. A system, comprising:
a midplane board;
a first sled comprising a first processor and a first interconnect part, the first interconnect part being coupled with the first processor, disposed on a lateral edge of the first sled, and received by a first interconnect portion disposed on one side of the midplane board;
a second sled comprising a second processor and a second interconnect part, the second interconnect part being coupled with the second processor, disposed on a lateral edge of the second sled, and received by a second interconnect portion disposed on the one side of the midplane board;
a third sled comprising a third processor and a third interconnect part, the third interconnect part being coupled with the third processor, disposed on a lateral edge of the third sled, and received by a third interconnect portion disposed on an other side of the midplane board;
a fourth sled comprising a fourth processor and a fourth interconnect part, the fourth interconnect part being coupled with the fourth processor, disposed on a lateral edge of the fourth sled, and received by a fourth interconnect portion disposed on the other side of the midplane board; and
a power harness comprising respective power connectors configured to be connected to respective rear edges of the first sled, second sled, third sled, and fourth sled, wherein at least one of the power connectors is configured to receive in a longitudinal direction to establish connection for transmission of power and to permit translation therein substantially perpendicular to the longitudinal direction;
wherein the first, second, third, and fourth interconnect portions of the midplane board are connected so as to enable inter-processor communication among the first, second, third, and fourth processors.

7. The system of claim 6 , wherein the midplane board is configured to receive at least eight sleds and enable inter-processor communication among processors of the at least eight sleds.

8. The system of claim 6 , wherein the first sled, the second sled, the third sled, and the fourth sled are replicas of each other.

9. The system of claim 8 , wherein the first sled on the one side of the midplane board is arranged in an orientation that is flipped 180 degrees from an orientation of the replica third sled on the other side of the midplane board.

10. The system of claim 6 , wherein the midplane board comprises a first midplane board, the first midplane board is in a server rack chassis, and the system further comprises:
a second midplane board in the server rack chassis; and
two additional sleds coupled with one side of the second midplane board and two additional sleds coupled with an other side of the second midplane board, wherein the second midplane board comprises connections to enable inter-processor communication among processors within the sleds coupled with the second midplane board.

11. The system of claim 6 , further comprising:
a server rack chassis containing the midplane board.

12. The system of claim 6 , wherein the first sled, second sled, third sled, and fourth sled are configured to be installed by a first motion for connecting to power by moving in a longitudinal direction of a server rack chassis containing the midplane board, and by a second motion for connecting for data transfer by translating toward the midplane board and non-parallel to the longitudinal direction.

13. The system of claim 6 , wherein the first interconnect part and the first interconnect portion comprise respective components configured for an Ultra Path Interconnect (UPI) interface.

14. The system of claim 6 , wherein the first interconnect part and the first interconnect portion comprise respective components configured for a Peripheral Component Interconnect Express (PCIE) interface.

15. A method comprising:
engaging a first rear end of a first sled with a first power connector within a server rack chassis by translating the first sled in a longitudinal direction of the server rack chassis; and
subsequent to the engaging of the first rear end of the first sled with the first power connector, engaging a first interconnect part on a first lateral edge of the first sled with a first interconnect portion on one side of a midplane board within the server rack chassis by translating the first sled toward the midplane board and non-parallel to the longitudinal direction.

16. The method of claim 15 , further comprising: connecting a first processor on the first sled for inter-processor communication by engaging a second sled, a third

Assignees

• Amazon Tech Inc

Inventors

• Rivnay Roey

Classifications

• H05K7/1492 Primary

  having electrical distribution arrangements, e.g. power supply or data communications · CPC title

• H05K7/1487

  Blade assemblies, e.g. blade cases or inner arrangements within a blade · CPC title

• H05K7/1491 Primary

  having cable management arrangements (management of optical cables G02B6/444; in telecommunication cabinets H04Q1/06) · CPC title

• H05K7/1445

  with double-sided connections · CPC title

• H05K7/1451

  with connections between circuit boards or units · CPC title