PCI express fabric routing for a fully-connected mesh topology

1. A PCIe Fabric comprising:

an Input/Output (I/O) tier comprising a first IO tier switch;

a hub tier comprising a plurality of hub tier switches, wherein the plurality of hub tier switches comprises a first hub tier switch and a second hub tier switch, wherein the IO tier is connected to the hub tier;

wherein the first IO tier switch is configured to:

receive, from a client, a transaction layer packet (TLP);

make a first determination that an address in the TLP is not associated with any multicast address range in the first IO tier switch;

make a second determination that the address in TLP is not associated with any downstream port in the first IO tier switch;

based on the first and second determinations, route the TLP to the first hub tier switch using a first upstream port on the first IO tier switch;

wherein the first hub tier switch is configured to:

receive, from the first IO tier switch, the TLP at a first downstream port;

make a third determination that the TLP is associated with a first multicast group;

based on the third determination:

generate a first rewritten TLP and route the first rewritten TLP to a first target device via a second downstream port on the first hub tier switch, wherein the second downstream port is a subscriber of the first multicast group.

2. The PCIe fabric of claim 1 , wherein the first hub tier switch is further configured to:

generate a second rewritten TLP and route the second rewritten TLP to a second target device via a third downstream port on the first hub tier switch, wherein the third downstream port is a second subscriber of the first multicast group.

3. The PCIe fabric of claim 1 , further comprising:

a processor, wherein the processor is connected to the first hub tier switch using a link between a second upstream port on the first hub tier switch and a root port on the processor,

wherein the first hub tier switch is further configured to:

based on the third determination, generate a second rewritten TLP and route the second rewritten TLP to the processor via the second upstream port on the first hub tier switch, wherein the second upstream port is a third subscriber of the first multicast group.

4. The PCIe fabric of claim 3 , further comprising:

a second processor, wherein the processor and the second processor are operatively connected via a processor interconnect.

5. The PCIe Fabric of claim 1 ,

wherein the first IO tier switch is further configured to:

receive, from the client, a second TLP;

make a fourth determination that the second TLP comprises an address that is within an address range associated with a third downstream port on the first IO tier switch;

based on the fourth determination, route the second TLP to the second hub tier switch using a crosslink;

wherein the second hub tier switch is configured to:

receive, from the first IO tier switch, the second TLP;

make a fifth determination that the second TLP is associated with a second target device directly connected to the second hub tier switch;

based on the fifth determination, route the second TLP to the second target device directly connected to the second hub tier switch.

6. The PCIe Fabric of claim 1 ,

wherein the first IO tier switch is further configured to:

receive, from the client, a second TLP;

make a fourth determination that the second TLP comprises an address that is within an address range associated with a third downstream port on the first IO tier switch;

based on the fourth determination, route the second TLP to the second hub tier switch using a crosslink;

wherein the second hub tier switch is configured to:

receive, from the first IO tier switch, the second TLP;

make a fifth determination that the second TLP is associated a second multicast group:

based on the fifth determination, generate a second rewritten TLP and route the second rewritten TLP to a second target device via a fourth downstream port on the second hub tier switch, wherein the fourth downstream port is a subscriber of the second multicast group, and wherein the second target device is connected to the second hub tier switch.

7. The PCIe Fabric of claim 1 ,

wherein the first IO tier switch is further configured to:

receive, from the client, a second TLP;

make a fourth determination that the second TLP is associated with the first multicast group;

based on the fourth determination, generate a second rewritten TLP and route the second rewritten TLP to the first hub tier switch;

wherein the first hub tier switch is configured to:

receive, from the first IO tier switch, the second rewritten TLP at the first downstream port;

make a fifth determination that the second rewritten TLP is associated with the target device directly connected to the first hub tier switch;

based on the fifth determination, route the second rewritten TLP to the first target device directly connected to the second hub tier switch.

8. The PCIe fabric of claim 1 , wherein the first upstream port on the first IO tier switch is associated with a first address range that includes all memory in the PCIe fabric excluding memory directly connected to a processor and a second address range that includes all memory associated with clients connected to the PCIe fabric, wherein the first address range and the second address range are used to route TLP packets received by the first IO tier switch.

9. The PCIe fabric of claim 8 , wherein all the memory associated with the clients comprises at least one selected from a group consisting of persistent storage and volatile memory.

10. The PCIe fabric of claim 8 , wherein all the memory in the PCIe fabric comprises at least one selected from a group consisting of persistent storage and volatile memory.

11. The PCIe fabric of claim 1 , wherein the first upstream port on the first hub tier switch is associated with a first address range that includes all memory in all slices in the PCIe fabric and a second address range that includes all memory associated with clients connected to the PCIe fabric, wherein the first IO tier switch and the first hub tier switch are in a first slice of the slices in the PCIe fabric, wherein the first slice excludes memory directly connected to the processor and excludes any of a plurality of target devices that are not directly connected to the first hub tier switch, wherein the first address range and the second address range are used to route TLP packets received by the first IO tier switch.

12. The PCIe fabric of claim 1 , wherein a crosslink connects the first IO tier switch and the second hub tier switch.

13. The PCIe fabric of claim 12 , wherein a downstream port of the first IO tier switch that is connected to the crosslink is associated with an address range that includes memory in a slice of the PCIe fabric, wherein the memory in the slice of the PCIe fabric includes a plurality of the target devices that are directly connected to the second hub tier switch and excludes any of the plurality of the target devices that are directly connected to the first hub tier switch.

14. The PCIe fabric of claim 1 , wherein first target device comprises at least one selected from a group consisting of volatile memory and persistent storage.

15. The PCIe fabric of claim 1 , wherein first target device comprises NAND flash.

16. The PCIe fabric of claim 1 , wherein first target device is one selected from a group consisting of a network endpoint device and a storage endpoint device.

17. The PCIe fabric of claim 1 , wherein the client is a PCIe endpoint.

18. The PCIe fabric of claim 1 , wherein the first IO tier switch and the first hub tier switch are connected using a crosslink.

19. The PCIe fabric of claim 18 , wherein the first IO tier switch is configured to route TLPs to the first hub tier switch using the crosslink.

20. The PCIe fabric of claim 18 , wherein the crosslink is not associated with any multicast group.

21. The PCIe fabric of claim 18 , wherein the crosslink is connected to a downstream port on the first IO tier switch and a third downstream port on the first hub tier switch.