Reliable communication between multi-processor clusters of multi-cluster computer systems
View Patent ↗Improved techniques are provided for detecting and correcting errors and skew in inter-cluster communications within computer systems having a plurality of multi-processor clusters. The local nodes of each cluster include a plurality of processors and an interconnection controller. Intra-cluster links are formed between the local nodes, including the interconnection controller, within a cluster. Inter-cluster links are formed between interconnection controllers of different clusters. Intra-cluster packets may be serialized and encapsulated as inter-cluster packets for transmission on inter-cluster links, preferably with link-layer encapsulation. Each inter-cluster packet may include a sequence identifier and error information computed for that packet. Clock data may be embedded in symbols sent on each bit lane of the inter-cluster links. Copies of transmitted inter-cluster packets may be stored until an acknowledgement is received. The use of inter-cluster packets on an inter-cluster link is preferably transparent to other links and to the protocol layer.
1. A computer-implemented method for detecting errors in a computer system comprising a plurality of clusters, each cluster including a plurality of local nodes and an interconnection controller interconnected by point-to-point intra-cluster links, communications between the local nodes and the interconnection controller made via an intra-cluster protocol using intra-cluster packets, the interconnection controller of each cluster interconnected by point-to-point inter-cluster links with the interconnection controller of other clusters, the computer-implemented method comprising:
forming an inter-cluster packet by encapsulating an intra-cluster packet;
encoding a sequence identifier in the inter-cluster packet;
calculating first cyclic redundancy code check data based only upon the inter-cluster packet;
encoding the first cyclic redundancy code check data in the inter-cluster packet; and
transmitting the inter-cluster packet from a first cluster to a second cluster on a point-to-point inter-cluster link.
2. The computer-implemented method of claim 1 , wherein the encoding steps comprise encoding in an area of the inter-cluster packet reserved for link layer information.
3. The computer-implemented method of claim 1 , further comprising:
receiving the inter-cluster packet; and
calculating second cyclic redundancy code check data based only upon the inter-cluster packet.
4. The computer-implemented method of claim 3 , further comprising:
detecting an error in the inter-cluster packet based upon the second cyclic redundancy code check data; and
notifying the first cluster of the error.
5. An apparatus for detecting errors in a computer system comprising a plurality of clusters, each cluster including a plurality of local nodes and an interconnection controller interconnected by point-to-point intra-cluster links, communications between the local nodes and the interconnection controller made via an intra-cluster protocol using intra-cluster packets, the interconnection controller of each cluster interconnected by point-to-point inter-cluster links with the interconnection controller of other clusters, the apparatus comprising:
means for forming an inter-cluster packet by encapsulating an intra-cluster packet;
means for encoding a sequence identifier in the inter-cluster packet;
means for calculating first cyclic redundancy code check data based only upon the inter-cluster packet;
means for encoding the first cyclic redundancy code check data in the inter-cluster packet; and
means for transmitting the inter-cluster packet from a first interconnection controller to a second interconnection controller on a point-to-point inter-cluster link.
6. A computer system, comprising:
a first cluster including a first plurality of processors and a first interconnection controller, the first plurality of processors and the first interconnection controller interconnected by first point-to-point intra-cluster links; and
a second cluster including a second plurality of processors and a second interconnection controller, the second plurality of processors and the second interconnection controller interconnected by second point-to-point intra-cluster links, the first interconnection controller coupled to the second interconnection controller by point-to-point inter-cluster links, communications on the first and second intra-cluster links made via an intra-cluster protocol by intra-cluster packets;
wherein the first interconnection controller is configured to:
receive an intra-cluster packet from a first processor in the first plurality of processors;
store the intra-cluster packet in a buffer;
add a header, including a sequence identifier, to the intra-cluster packet to form a high-speed link packet;
compute a first cyclic redundancy code check based only upon the high-speed link packet;
encode first cyclic redundancy code check data in the high-speed link packet; and
transmit the high-speed link packet to the second interconnection controller in the second cluster;
wherein the second interconnection controller is configured to:
receive the high-speed link packet;
compute a second cyclic redundancy code check based only upon the high-speed link packet;
compare results of the second cyclic redundancy code check with the encoded first cyclic redundancy code check data in the high-speed link packet; and
notify the first interconnection controller regarding the results of the comparison.
7. The computer system of claim 6 , wherein the second interconnection controller is further configured to detect a gap in sequence identifiers of high-speed link packets received from the first interconnection controller, and wherein the notifying step comprises notifying the first interconnection controller of the gap.
8. The computer system of claim 6 , wherein the notifying step comprises sending an ACK packet to the first interconnection controller indicating a sequence identifier of one or more high-speed link packets received without errors.
9. The computer system of claim 8 , wherein the first interconnection controller is further configured to purge intra-cluster packets stored in the buffer that correspond to the sequence identifiers of the ACK packets.
10. The computer system of claim 6 , wherein the notifying step comprises sending a NACK packet to the first interconnection controller indicating a sequence identifier of one or more high-speed link packets received with errors.
11. The computer system of claim 6 , wherein the notifying step comprises initiating a retry sequence when an error is detected in the comparing step, the retry sequence causing the first interconnection controller to transmit high-speed link packets that include copies of intra-cluster packets stored in the buffer.
12. The computer system of claim 11 , further comprising the step of sending, after the retry sequence is complete, an acknowledgement packet with a sequence identifier of a last packet received without error.
13. The computer system of claim 11 , further comprising the step of sending, after the retry sequence is complete, an negative acknowledgement packet with a sequence identifier of a last packet received having an error.
14. The computer system of claim 11 , wherein the retry sequence provides a one-time, error-free delivery of protocol layer packets.
15. The computer system of claim 6 , wherein the notifying step comprises initiating a retry sequence when a gap is detected, the retry sequence causing the first interconnection controller to transmit high-speed link packets that include copies of intra-cluster packets stored in the buffer.
16. An interconnection controller, comprising:
an intra-cluster interface configured for coupling with intra-cluster links to a plurality of local processors arranged in a point-to-point architecture in a local cluster;
an inter-cluster interface configured for coupling with an inter-cluster link to a non-local interconnection controller in a non-local cluster;
a transceiver configured to:
receive an intra-cluster packet from a local processor via an intra-cluster link;
encode a sequence identifier in a header of the intra-cluster packet;
compute cyclic redundancy code check data based only on the encoded packet; and
encode the cyclic redundancy code check data in the encoded packet; and
a serializer/deserializer configured to serialize the encoded packet and forward the encoded, serialized packet to the inter-cluster interface for transmission to the non-local interconnection controller via an inter-cluster link.
17. The interconnection controller of claim 16 , wherein the inter-cluster interface is further configured to receive encoded, serialized packets from the non-local interconnection controller, wherein the serializer/deserializer is further configured to deserialize the encoded, serialized packets and wherein the transceiver is further configured to perform a cyclic redundancy code check on the deserialized packets.
18. An integrated circuit comprising the interconnection controller of claim 16 .
19. The integrated circuit of claim 18 , wherein the integrated circuit comprises an application-specific integrated circuit.
20. A set of semiconductor processing masks representative of at least a portion of the interconnection controller of claim 16 .
21. At least one computer-readable medium having data structures stored therein representative of the interconnection controller of claim 16 .
22. The at least one computer-readable medium of claim 21 , wherein the data structures comprise a simulatable representation of the interconnection controller.
23. The at least one computer-readable medium of claim 22 , wherein the simulatable representation comprises a netlist.
24. The at least one computer-readable medium of claim 21 , wherein the data structures comprise a code description of the interconnection controller.
25. The at least one computer-readable medium of claim 24 , wherein the code description corresponds to a hardware description language.