High performance, scalable multi chip interconnect
View Patent ↗A flexible, scalable server is described. The server includes plural server nodes each server node including processor cores and switching circuitry configured to couple the processor to a network among the cores with the plurality of cores implementing networking functions within the compute nodes wherein the plurality of cores networking capabilities allow the cores to connect to each other, and to offer a single interface to a network coupled to the server.
1. A server, comprising:
a plurality of tiled compute nodes that are arranged in a compute node stack, with multiple ones of the plurality of tiled compute nodes comprising:
a processor;
memory devices coupled to the processor; and
a layer 2 switch circuitry configured to couple the processor of the tiled compute node to a network among the processors of the tiled compute nodes, and which, with the compute node stack, provides a single interface to the network; and with the processor of each compute node in the server configured to:
execute an operating system;
provide network switching functions within the respective plurality of tiled compute nodes of the server; and
external memory devices that are external to the tiled compute nodes and accessed by the tiled compute nodes.
2. The server of claim 1 , wherein the processors are further configured to:
connect tiled compute nodes to other tiled compute nodes to provide the networking switching functions within the single interface to the network coupled to the server.
3. The server of claim 1 , wherein the layer 2 switch circuity uses media access control addresses for forwarding frames.
4. The server of claim 1 , wherein when adding additional tiled compute nodes, the additional tiled compute nodes are added without adding additional external networking components to the server.
5. A server comprising:
plural server compute nodes, each server compute node comprising:
external memory devices;
a plurality of processor tiles coupled to the external memory devices, the plurality of processor tiles arranged in a stack topology, with multiple ones of the plural processor cores each comprising:
a processor; and
layer 2 switching circuitry configured to couple the processor to a network among the plural processor cores; and with the multiple ones of the processor cores of the plurality of processor tiles further configured to provide:
network switching within the server compute nodes by the plural processor cores; and
distributed networking hardware services including one or more of packet classification, load balancing, buffer management and packet egress-reordering, with, for each respective server compute node, the plural processor cores are connected over a network that provides a single interface to an external network that couples to the respective server compute node.
6. The server of claim 5 wherein the networking functions in the server compute nodes further comprise deep packet inspection.
7. The server of claim 5 wherein the hardware services includes the load balancing, and the server compute node executes software running on the processor cores or in firmware in that server compute node to provide the load balancing across multiple server compute nodes with a set of server compute nodes or processor cores providing a single network endpoint.
8. The server of claim 7 wherein routing is implemented in fast-path software running within the processor cores supporting layer 2 routing to IP routing, load-based routing, or “address” based routing.
9. The server of claim 7 further comprising:
fast path hardware in at least one processor core of the processor cores to allow low-latency and high-bandwidth switching capability for that at least one processor core.
10. The server of claim 7 wherein routing functions are arranged with processor cores executing control path algorithms, and with data path algorithms managed by hardware.
11. The server of claim 10 wherein the server further includes dedicated hardware to which the hardware services are offloaded without using the plurality of processor tiles.
12. The server of claim 10 wherein local traffic is identified based on a physical port number or MAC address or system load information.
13. The server of claim 5 wherein a packet is switched based on a range comparison with an address field in the packet.
14. A method of operating a set of server nodes that comprise plural multi-core processor chips, the method comprising:
arranging the set of server nodes in a node stacking topology with each server node including a L2 (layer 2) switch to provide scalable node-stacking;
receiving packets that arrive at a receiver interface in a multi-core processor chip for a given one of the set of server nodes;
parsing by a classifier engine executed by the multi-core processor chip, the received packets to produce packet descriptors that determine a destination to which the received packets belongs with either the classifier engine or a processor core of the multi-core processor chip determining an egress port for the received packets;
applying by the processor core a load balancing algorithm to determine a processor core to process the packet; and
enqueueing by the multi-core processor chip the packet descriptors generated by the classifier.
15. The method of claim 14 , wherein applying load balancing, further comprises:
executing software on the processor cores of the multi-core processor to provide load balancing across multiple ones of the server nodes, with the server nodes or processors providing a single network endpoint.
16. The method of claim 14 , further comprising:
providing hardware services that include load balancing across multiple server nodes with a set of server nodes or processors cores providing a single network endpoint.