Granted Patent B2

US 8,553,683 · App. 13/176,350 · Granted Oct 8, 2013

Abstract

Claims (32)

1. A method for operating an intra-domain switch in a first set of intra-domain switches in a non-blocking switch fabric having the first set of intra-domain switches, a second set of intra-domain switches, and a set of inter-domain switches located between the first set of intra-domain switches and the second set of intra-domain switches, the method comprising:

at an outside edge port of the intra-domain switch, inserting, in a packet that must travel between domains, an inter-domain routing prefix identifying both the source and destination domains; and

passing the packet that must travel between domains to one of the inter-domain switches via an inside edge port of the intra-domain switch, the method further comprising:

classifying an incoming packet as an ordered packet or an unordered packet;

for an ordered packet, forwarding the incoming packet to a port identified by a spread routing table using a hash of source and destination identifications as a table index; and

for an unordered packet, forwarding the incoming packet to a currently least utilized port, that leads to a root of the non-blocking switch fabric, having a smallest accumulator value, wherein an accumulator value for a port is increased by a value proportional to a length of each packet being forwarded to that port.

2. The method of claim 1 , further comprising:

determining if the packet is a read request or other non-posted request;

when the packet is a read request or other non-posted request, saving a source domain identification from a prefix of the packet for use in creating a prefix to return route an eventual returning completion for the packet.

3. The method of claim 2 , further comprising:

at the outside edge port of the intra-domain switch, removing, from a packet received via an intra-domain switch, an inter-domain routing prefix identifying both source and destination domains and saving the source domain to be used to route a response packet back to the original source port.

4. The method of claim 2 , wherein the inserting is performed by non-transparent bridge logic functionality.

5. The method of claim 2 , wherein the inserting is performed by a direct memory access (DMA) controller or other form of message passing engine.

6. The method of claim 1 , wherein an unordered packet includes a completion packet.

7. The method of claim 2 , further comprising:

when a fault is known to be present in a destination domain, performing a route exception lookup in a route exception lookup table at each step leading towards a root of a switch fabric defined by the first set of intra-domain switches, second set of intra-domain switches, and set of inter-domain switches; and

prohibiting the packet from being forwarded to a destination port identified in the route exception lookup table.

8. The method of claim 7 , wherein the fault is known to be present by detecting a tag in the packet having a bit that indicates that the destination domain has a fault.

9. The method of claim 7 , wherein inside edge switches of each source domain are configured with status bits for each egress port connecting to an inter-domain switch, and wherein if the inter-domain switch or the links to it contain a fault, a route exception lookup is performed using the destination domain.

10. The method of claim 9 , wherein if a route exception is indicated by the lookup, then for ordered traffic, a next higher numbered port with good status is selected, and for unordered traffic, all ports with bad status are excluded from deficit round robin route choices.

11. A non-transitory program storage device readable by a machine tangibly embodying a program of instructions executable by the machine to perform a method for operating an intra-domain switch in a first set of intra-domain switches in a non-blocking switch fabric having the first set of intra-domain switches, a second set of intra-domain switches, and a set of inter-domain switches located between the first set of intra-domain switches and the second set of intra-domain switches, the method comprising:

at an outside edge port of the intra-domain switch, inserting, in a packet that must travel between domains, an inter-domain routing prefix identifying both a source and a destination domain; and

passing the packet that must travel between domains to one of the inter-domain switches via an inside edge port of the intra-domain switch, the method including:

classifying an incoming packet as an ordered packet or an unordered packet;

for an ordered packet, forwarding the incoming packet to a port identified by a spread routing table using a hash of source and destination identifications as a table index; and

for an unordered packet, forwarding the incoming packet to a currently least utilized port, that leads to a root of the non-blocking switch fabric, having a smallest accumulator value, wherein an accumulator value for a port is increased by a value proportional to a length of each packet being forwarded to that port.

12. An apparatus operating an intra-domain switch in a first set of intra-domain switches in a non-blocking switch fabric having the first set of intra-domain switches, a second set of intra-domain switches, and a set of inter-domain switches located between the first set of intra-domain switches and the second set of intra-domain switches, the apparatus comprising:

means for, at an outside edge port of the intra-domain switch, inserting, in a packet that must travel between domains, an inter-domain routing prefix identifying both source destination domains;

means for passing the packet that must travel between domains to one of the intra-domain switches via an inside edge port of the intra-domain switch,

means for classifying an incoming packet as an ordered packet or an unordered packet;

for an ordered packet, forwarding the incoming packet to a port identified by a spread routing table using a hash of source and destination identifications as a table index; and

for an unordered packet, forwarding the incoming packet to a currently least utilized port, that leads to a root of the non-blocking switch fabric, having a smallest accumulator value, wherein an accumulator value for a port is increased by a value proportional to a length of each packet being forwarded to that port.

Assignments (9)