Abstract: Systems and method for routing data packets in ring network. A data packet being transmitted to a destination node may be received by a first structure at a first node. The first node may determine a number of hops the data packet will traverse as it is transmitted from the first node to the destination node and compare the determined number of hops to a threshold hop value to determine whether the number of hops is equal to or less than the threshold hop value. If the number of hops is greater than the threshold, the data packet may be transmitted to a dimension queuing structure for a first virtual channel within a second node, otherwise, the data packet may be transmitted to a dimension queuing structure for a second virtual channel or a turn queuing structure within the second node.

Abstract: Generally disclosed herein is an approach for optimizing routing strategy to tolerate faults in a toroidal network topology including, but not limited to, N-dimensional mesh, torus, and twisted torus. The approach may include balancing a load for a specified input traffic pattern operating offline or online. The approach may also include an optimization enhancement technique specifically applicable to symmetric, dynamically composable toroidal networks based on a set of centrally connected circuit switches.

Abstract: Methods, systems, and apparatus, including computer-readable storage media for hybrid network topologies that flexibly support diverting bandwidth for indirect connections to bandwidth for direct connections for computing devices arranged in the hybrid network topology, on a workload-by-workload basis. The hybrid network topology described herein arranges computing devices into groups and connects neighboring computing devices to common switches based on the dimensionality of the computing device groups. Indirect connections connecting devices in different groups through a common switch can be at least partially diverted to form an additional direct connection between the neighboring computing devices in the same group. The network of nodes can divert bandwidth, for example based on executing different workloads with different bandwidth requirements for executing the workload locally, e.g., within a group of nodes, versus globally, e.g., across different groups of nodes.

Abstract: Aspects of the disclosure are directed to a randomly determined first hop for network routing in an N-dimensional grid network. The first hop for a packet traversing the network is randomly determined based on a probability distribution for potential first hops. Based on a destination for the packet, a router in the network can search for potential paths and corresponding probabilities for the first hop, select a path based on the first hop, and forward the packet along that path to its destination. Network routing with a randomly determined first hop can improve fault tolerance without adding network complexities to the N-dimensional grid network.

Abstract: The technology generally relates to systems and methods for operating a distributed processing network. Tenser Processing Units (TPUs) may be interconnected to one another as part of a TPU cluster or pod. A number of TPU clusters may be connected to one another via inter-cluster switches to form a distributed network of TPU clusters. The inter-cluster switches may be configured to efficiently direct data transmissions from a first TPU cluster to a second TPU cluster based on identification of the data transmission's final destination and performing next-hop lookup operations. The inter-cluster switches may also be configured to perform cut-through operations and may be configured to have an internal non-blocking connectivity.

Abstract: Systems and method for routing data packets in ring network. A data packet being transmitted to a destination node may be received by a first structure at a first node. The first node may determine a number of hops the data packet will traverse as it is transmitted from the first node to the destination node and compare the determined number of hops to a threshold hop value to determine whether the number of hops is equal to or less than the threshold hop value. If the number of hops is greater than the threshold, the data packet may be transmitted to a dimension queuing structure for a first virtual channel within a second node, otherwise, the data packet may be transmitted to a dimension queuing structure for a second virtual channel or a turn queuing structure within the second node.

Abstract: Systems and method for routing data packets in ring network. A data packet being transmitted to a destination node may be received by a first structure at a first node. The first node may determine a number of hops the data packet will traverse as it is transmitted from the first node to the destination node and compare the determined number of hops to a threshold hop value to determine whether the number of hops is equal to or less than the threshold hop value. If the number of hops is greater than the threshold, the data packet may be transmitted to a dimension queuing structure for a first virtual channel within a second node, otherwise, the data packet may be transmitted to a dimension queuing structure for a second virtual channel or a turn queuing structure within the second node.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer-storage media, for connecting processors using twisted torus configurations. In some implementations, a cluster of processing nodes is coupled using a reconfigurable interconnect fabric. The system determines a number of processing nodes to allocate as a network within the cluster and a topology for the network. The system selects an interconnection scheme for the network, where the interconnection scheme is selected from a group that includes at least a torus interconnection scheme and a twisted torus interconnection scheme. The system allocates the determined number of processing nodes of the cluster in the determined topology, sets the reconfigurable interconnect fabric to provide the selected interconnection scheme for the processing nodes in the network, and provides access to the network for performing a computing task.

Abstract: Generally disclosed herein is an approach for optimizing routing strategy to tolerate faults in a toroidal network topology including, but not limited to, N-dimensional mesh, torus, and twisted torus. The approach may include balancing a load for a specified input traffic pattern operating offline or online. The approach may also include an optimization enhancement technique specifically applicable to symmetric, dynamically composable toroidal networks based on a set of centrally connected circuit switches.

Abstract: Improvements in a molecular-dynamic simulator provide ways to save energy during computation and reduce die area consumed on an integrated circuit. Examples of such improvements include different interaction modules for different ranges, the use of streaming along rows while multicasting along columns in an array of interaction modules, the selection of computation units based on balancing computational costs and communication costs, the use of fences in networks that connect computation units, and the use of bond calculators to carry out specialized bond calculations.

Abstract: Systems and method for routing data packets in ring network. A data packet being transmitted to a destination node may be received by a first structure at a first node. The first node may determine a number of hops the data packet will traverse as it is transmitted from the first node to the destination node and compare the determined number of hops to a threshold hop value to determine whether the number of hops is equal to or less than the threshold hop value. If the number of hops is greater than the threshold, the data packet may be transmitted to a dimension queuing structure for a first virtual channel within a second node, otherwise, the data packet may be transmitted to a dimension queuing structure for a second virtual channel or a turn queuing structure within the second node.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer-storage media, for connecting processors using twisted torus configurations. In some implementations, a cluster of processing nodes is coupled using a reconfigurable interconnect fabric. The system determines a number of processing nodes to allocate as a network within the cluster and a topology for the network. The system selects an interconnection scheme for the network, where the interconnection scheme is selected from a group that includes at least a torus interconnection scheme and a twisted torus interconnection scheme. The system allocates the determined number of processing nodes of the cluster in the determined topology, sets the reconfigurable interconnect fabric to provide the selected interconnection scheme for the processing nodes in the network, and provides access to the network for performing a computing task.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer-storage media, for connecting processors using twisted torus configurations. In some implementations, a cluster of processing nodes is coupled using a reconfigurable interconnect fabric. The system determines a number of processing nodes to allocate as a network within the cluster and a topology for the network. The system selects an interconnection scheme for the network, where the interconnection scheme is selected from a group that includes at least a torus interconnection scheme and a twisted torus interconnection scheme. The system allocates the determined number of processing nodes of the cluster in the determined topology, sets the reconfigurable interconnect fabric to provide the selected interconnection scheme for the processing nodes in the network, and provides access to the network for performing a computing task.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer-storage media, for connecting processors using twisted torus configurations. In some implementations, a cluster of processing nodes is coupled using a reconfigurable interconnect fabric. The system determines a number of processing nodes to allocate as a network within the cluster and a topology for the network. The system selects an interconnection scheme for the network, where the interconnection scheme is selected from a group that includes at least a torus interconnection scheme and a twisted torus interconnection scheme. The system allocates the determined number of processing nodes of the cluster in the determined topology, sets the reconfigurable interconnect fabric to provide the selected interconnection scheme for the processing nodes in the network, and provides access to the network for performing a computing task.

Abstract: A data communication apparatus includes a router, first and second packet producers, and a penalizer. The router is directly connected to the first and second producers. The penalizer assesses penalties against each producer whenever that producer is serviced. The penalty value depends at least in part on an expected extent to which the first producer requires service. The penalizer then accumulates penalties against each producer.

Abstract: A data communication apparatus includes a router, first and second packet producers, and a penalizer. The router is directly connected to the first and second producers. The penalizer assesses penalties against each producer whenever that producer is serviced. The penalty value depends at least in part on an expected extent to which the first producer requires service. The penalizer then accumulates penalties against each producer.

Abstract: An approach to data communication makes use of a protocol for encoding data on a serial link that provides both a run length limiting function and a frame marking function, while minimizing communication overhead over the data bearing portions of the signal, and while limiting latency introduced into the communication. In some examples, a single bit is added as a frame marker in such a way that a single bit frame marker also limits run length.

Abstract: An approach to introducing adaptive routing into a communication approach for passing messages between nodes over links between the nodes includes forming virtual channels over the links of the system and defining a deterministic routing function over the virtual channels such that the deterministic routing function is deadlock free. Adaptive routing is then permitted at nodes using the existing virtual channels by introducing a constraint on the available virtual channels used to forward communication that arrives at a node for a particular destination. The constraint on the virtual channels is such that the adaptive system is also deadlock free.

Abstract: An approach to data communication makes use of a protocol for encoding data on a serial link that provides both a run length limiting function and a frame marking function, while minimizing communication overhead over the data bearing portions of the signal, and while limiting latency introduced into the communication. In some examples, a single bit is added as a frame marker in such a way that a single bit frame marker also limits run length.

Abstract: An approach to introducing adaptive routing into a communication approach for passing messages between nodes over links between the nodes includes forming virtual channels over the links of the system and defining a deterministic routing function over the virtual channels such that the deterministic routing function is deadlock free. Adaptive routing is then permitted at nodes using the existing virtual channels by introducing a constraint on the available virtual channels used to forward communication that arrives at a node for a particular destination. The constraint on the virtual channels is such that the adaptive system is also deadlock free.