Abstract: Provided is an apparatus for accelerating graph neural network (GNN) pre-processing, the apparatus including a set-partitioning accelerator configured to sort each edge of an original graph stored in a coordinate list (COO) format by a node number, perform radix sorting based on a vertex identification (VID) to generate a COO array of a preset length, and perform uniform random sampling on some nodes of a given node array, a merger configured to merge the COO array of the preset length to generate one sorted COO array, a re-indexer configured to assign new consecutive VIDs respectively to the nodes selected through the uniform random sampling, and a compressed sparse row (CSR) converter configured to the edges sorted by the node number into a CSR format.

Abstract: A memory expander is disclosed. The memory expander includes a memory, a memory controller configured to control the memory, a compute express link (CXL) engine configured to acquire a CXL flit from a host device connected to the memory expander and configured to acquire a calculation request for pieces of data stored in the memory by performing conversion on the CXL flit, and a domain-specific accelerator configured to perform a calculation in response to the calculation request.

Abstract: A computational storage supporting graph machine learning acceleration includes a solid state drive (SSD) configured to store a graph data set; and a field-programmable gate array (FPGA) configured to download, to a memory, a graph machine learning model programmed in a form of a data flow graph by a host, wherein a hardware logic built in the FPGA performs access to the SSD through a peripheral component interconnect-express (PCIe) switch.

Abstract: Provided is an apparatus for accelerating graph neural network (GNN) pre-processing, the apparatus including a set-partitioning accelerator configured to sort each edge of an original graph stored in a coordinate list (COO) format by a node number, perform radix sorting based on a vertex identification (VID) to generate a COO array of a preset length, and perform uniform random sampling on some nodes of a given node array, a merger configured to merge the COO array of the preset length to generate one sorted COO array, a re-indexer configured to assign new consecutive VIDs respectively to the nodes selected through the uniform random sampling, and a compressed sparse row (CSR) converter configured to the edges sorted by the node number into a CSR format.

Abstract: Provided is an apparatus for accelerating graph neural network (GNN) pre-processing, the apparatus including a conversion unit configured to convert an original graph in a coordinate list (COO) format into a graph in a compressed sparse row (CSR) format, a sub-graph generation unit configured to generate a sub-graph with a reduced degree of the graph in the CSR format, and an embedding table generation unit configured to generate an embedding table corresponding to the sub-graph.

Abstract: A compute express link (CXL) computing system includes a host device including a CPU that supports CXL, and a CXL storage connected to a CXL root port of the CPU based on the CXL interconnect and including a flash memory-based memory module.

Abstract: Provided is an apparatus for accelerating a graph neural network for efficient parallel processing of massive graph datasets, including a streaming multiprocess (SM) scheduler and a computation unit, wherein the SM scheduler obtains a subgraph and an embedding table per layer, determines a number of SMs to be allocated for processing embeddings of a destination-vertex based on a feature dimension and a maximum number of threads in each of the SMs, and allocates the determined number of SMs to each of all destination-vertices included in the subgraph, and the computation unit obtains, by each of the SMs, embeddings of a destination-vertex allocated to each SM, obtains, by each SM, embeddings of at least one or more neighbor-vertices of the destination-vertex using the subgraph, and performs, by each SM, a user-designated operation using the embeddings of the destination-vertex and the embeddings of the neighbor-vertices.

Abstract: A memory controlling device configured to connect to a memory module including a resistance switching memory cell array which is partitioned into a plurality of partitions including a first partition and a second partition is provided. A first controlling module accesses the memory module. A second controlling module determines whether there is a conflict for the first partition to which a read request targets when an incoming request is the read request, instructs the first controlling module to read target data of the read request from the memory module when a write to the second partition is in progress, and suspends the read request when a write to the first partition is in progress.

Abstract: A computational storage supporting graph machine learning acceleration includes a solid state drive (SSD) configured to store a graph data set; and a field-programmable gate array (FPGA) configured to download, to a memory, a graph machine learning model programmed in a form of a data flow graph by a host, wherein a hardware logic built in the FPGA performs access to the SSD through a peripheral component interconnect-express (PCIe) switch.

Abstract: A first IP address is set to the host, and a second IP address is set to the storage card. A storage card includes a storage device and a processor for executing firmware. The host converts a first Ethernet packet including an ISP-related request and destined for the second IP address into an NVMe request according to an NVMe protocol, and transfers the NVMe request to the storage card. The firmware parses the NVMe request to perform the ISP-related request.

Abstract: A method of supporting persistence of a computing device is provided. The computing device performs a stop procedure upon power failure. In the stop procedure, the computing device schedules out a running process task, stores a state of the process task to a process control block of a memory module including a non-volatile memory, flushes a cache for the processor, and flushes a pending memory request.

Abstract: A first IP address is set to the host, and a second IP address is set to the storage card. A storage card includes a storage device and a processor for executing firmware. The host converts a first Ethernet packet including an ISP-related request and destined for the second IP address into an NVMe request according to an NVMe protocol, and transfers the NVMe request to the storage card. The firmware parses the NVMe request to perform the ISP-related request.

Abstract: A processor of the computing device includes a plurality of cores and executes one or more instructions stored in a memory module including a non-volatile memory, thereby performing a stop procedure upon a power failure and performing a go procedure upon power recovery. In the stop procedure, the processor accesses process control blocks of processes being run, scheduling each process to a run queue of a corresponding first core among first cores included in the cores, removes the scheduled process from the run queue and makes the removed process wait in a waiting queue, executes an idle task, and stops a device included in the computing device.

Abstract: A memory controlling device configured to connect to a memory module including a resistance switching memory cell array which is partitioned into a plurality of partitions including a first partition and a second partition is provided. A first controlling module accesses the memory module. A second controlling module determines whether there is a conflict for the first partition to which a read request targets when an incoming request is the read request, instructs the first controlling module to read target data of the read request from the memory module when a write to the second partition is in progress, and suspends the read request when a write to the first partition is in progress.

Abstract: A memory controlling device configured to connect to a memory module including a resistance switching memory cell array which is partitioned into a plurality of partitions including a first partition and a second partition is provided. A first controlling module accesses the memory module. A second controlling module determines whether there is a conflict for the first partition to which a read request targets when an incoming request is the read request, instructs the first controlling module to read target data of the read request from the memory module when a write to the second partition is in progress, and suspends the read request when a write to the first partition is in progress.

Abstract: A method of supporting persistence of a computing device is provided. The computing device performs a stop procedure upon power failure. In the stop procedure, the computing device schedules out a running process task, stores a state of the process task to a process control block of a memory module including a non-volatile memory, flushes a cache for the processor, and flushes a pending memory request.

Abstract: A memory controlling device of a computing device including a CPU, a memory, and a flash-based storage device is provided. The memory controlling device includes an address manager and an interface. The address manager aggregates a memory space of the memory and a storage space of the storage device into an expanded memory space, and handles a memory request for the expanded memory space from the CPU by using the memory space of the memory as a cache for the storage space of the storage device. The interface is used to access the memory and the storage device.

Abstract: A memory controlling device configured to connect to a memory module including a resistance switching memory cell array which is partitioned into a plurality of partitions including a first partition and a second partition is provided. A first controlling module accesses the memory module. A second controlling module determines whether there is a conflict for the first partition to which a read request targets when an incoming request is the read request, instructs the first controlling module to read target data of the read request from the memory module when a write to the second partition is in progress, and suspends the read request when a write to the first partition is in progress.

Abstract: A memory controlling device configured to connect to a first memory module including a resistance switching memory cell array which is partitioned into a plurality of partitions and a second memory module used for a cache is provided. A cache controller splits an address of a read request into at least a first cache index and a first tag, and determines whether the read request is a cache hit or a cache miss by referring to a lookup logic based on the first cache index and the first tag. The cache controller instructs the memory controller to read target data of the read request from the first memory module when the read request targets to the second partition in a case where the read request is the cache miss and a write to the first partition is in progress.

Abstract: A memory controlling device configured to connect to a first memory module including a resistance switching memory cell array which is partitioned into a plurality of partitions and a second memory module used for a cache is provided. A cache controller splits an address of a read request into at least a first cache index and a first tag, and determines whether the read request is a cache hit or a cache miss by referring to a lookup logic based on the first cache index and the first tag. The cache controller instructs the memory controller to read target data of the read request from the first memory module when the read request targets to the second partition in a case where the read request is the cache miss and a write to the first partition is in progress.