Abstract: A device for processing convolution operations includes: a processor that executes, in a neural network, a convolution operation on input data in a form of width×height×input channel and on a filter in a form of K×K×input channel or K×K to correspond to a form of the input data, K being an integer greater than or equal to one, and that generates output data in a form of width×height×output channel; and a reader that sequentially reads, from a memory storing the input data, a data group having more pieces of data than unit data throughput of an operator, and provides the data group to the operator to reuse at least one piece of data constituting the data group in the convolution operation. The processor executes, by using one or more operators identical to the operator, the convolution operation multiple times based on the unit data throughput.

Abstract: A method of operating a computing device includes a storage device receiving a request to execute an application instance, and executing the application instance at the storage device in response to the received request by the storage device. The application instance includes a plurality of storage instances connected with one another, and at least one of the plurality of storage instances is connected to a host device.

Abstract: Data deduplication is performed by separating data into a plurality of data chunks that correspond to first through Nth positions and include symbols, calculating discrimination indices of the positions using frequencies of the symbols in the different positions; arranging the order of the positions based on values of the discrimination indices; and generating fingerprints of the data through combination of data chunks that correspond to a number of the positions, based on the arranged order of the positions.

Abstract: The data storage device includes a separator configured to separate data requested to write by clients into data chunks, an address translator configured to translate first addresses generated by the data chunks into second addresses as global addresses, a storage node mapper configured to map the second addresses to a plurality of storage nodes, and a data store unit configured to select a target storage node among the plurality of storage nodes and store the data chunks in the target storage node. The data chunks include a plurality of data input/output unit blocks. If other data chunks that are the same with the data chunks are pre-stored in the plurality of storage nodes, the data store unit is configured to establish links between the same pre-stored data chunks and the second addresses, rather than stores the data chunks in the plurality of storage nodes.

Abstract: A method of operating a computing device includes a storage device receiving a request to execute an application instance, and executing the application instance at the storage device in response to the received request by the storage device. The application instance includes a plurality of storage instances connected with one another, and at least one of the plurality of storage instances is connected to a host device.

Abstract: A method includes determining a size of a recommended spare memory space of each of one or more storage nodes based on a state of the storage nodes, and adjusting a spare memory space of each of the storage nodes based on the size of the recommended spare memory space.

Abstract: A cloud-based sensor data acquisition system includes one or more sensors configured to acquire sensor data; a first terminal configured to transmit sensor data request information to a cloud server through an access point (AP); and one or more second terminals each configured to acquire sensor data from one or more sensors on a corresponding dissemination tree received from the cloud server; and transmit the acquired sensor data to the first terminal through the AP. The cloud server is configured to, in response to receiving the sensor data request information from the first terminal, build a corresponding dissemination tree for each of one or more second terminals within a sensing range; and transmit the corresponding dissemination tree to each of the one or more second terminals within the sensing range.

Abstract: Data deduplication is performed by separating data into a plurality of data chunks that correspond to first through Nth positions and include symbols, calculating discrimination indices of the positions using frequencies of the symbols in the different positions; arranging the order of the positions based on values of the discrimination indices; and generating fingerprints of the data through combination of data chunks that correspond to a number of the positions, based on the arranged order of the positions.

Abstract: A data deduplication method includes separating data into a plurality of data chunks that correspond to first to N-th positions, N being a positive integer that is greater than 1; determining discrimination indexes of the first to N-th positions, respectively; arranging the order of the first to N-th positions according to values of the discrimination indexes; recording the arranged order of the first to N-th positions on a position vector; and generating fingerprints through combination of the data chunks that correspond to the first to N-th positions according to the order of the first to N-th positions recorded on the position vector, wherein the determining discrimination indexes includes determining the discrimination indexes according to a ratio of duplicate data chunks to the data chunks that correspond to a same position in a plurality of pieces of data.

Abstract: The data storage device includes a separator configured to separate data requested to write by clients into data chunks, an address translator configured to translate first addresses generated by the data chunks into second addresses as global addresses, a storage node mapper configured to map the second addresses to a plurality of storage nodes, and a data store unit configured to select a target storage node among the plurality of storage nodes and store the data chunks in the target storage node. The data chunks include a plurality of data input/output unit blocks. If other data chunks that are the same with the data chunks are pre-stored in the plurality of storage nodes, the data store unit is configured to establish links between the same pre-stored data chunks and the second addresses, rather than stores the data chunks in the plurality of storage nodes.

Abstract: A hierarchical scheduling apparatus for a cloud environment includes a schedule configuring unit configured to classify a plurality of tasks into one or more local tasks and one or more remote tasks; a schedule delegating unit configured to transmit, to another resource, a list of the remote tasks and a list of available resources to delegate scheduling authority for the remote tasks to the other resource; and a scheduling unit configured to schedule the local tasks.

Abstract: A method includes determining a size of a recommended spare memory space of each of one or more storage nodes based on a state of the storage nodes, and adjusting a spare memory space of each of the storage nodes based on the size of the recommended spare memory space.

Abstract: A wear management apparatus and a wear management method of a storage system including storage nodes are provided. The wear management apparatus includes a monitor unit configured to collect status information about each of the storage nodes. The wear management apparatus further includes a wear management unit configured to establish a wear progress model with respect to the storage nodes based on the status information, and control a wear acceleration index of each of the storage nodes based on the wear progress model and a wear management policy.

Abstract: A hierarchical scheduling apparatus for a cloud environment includes a schedule configuring unit configured to classify a plurality of tasks into one or more local tasks and one or more remote tasks; a schedule delegating unit configured to transmit, to another resource, a list of the remote tasks and a list of available resources to delegate scheduling authority for the remote tasks to the other resource; and a scheduling unit configured to schedule the local tasks.

Abstract: Provided is a system for distributed multi-stage compilation of application programs in a cloud environment. The system includes a source apparatus to generate a compilation order in which source representation of an application program is compiled, and a destination apparatus to receive a compiled representation created by compiling at least a portion of the source representation according to the compilation order.

Abstract: A cloud-based sensor data acquisition system includes one or more sensors configured to acquire sensor data; a first terminal configured to transmit sensor data request information to a cloud server through an access point (AP); and one or more second terminals each configured to acquire sensor data from one or more sensors on a corresponding dissemination tree received from the cloud server; and transmit the acquired sensor data to the first terminal through the AP. The cloud server is configured to, in response to receiving the sensor data request information from the first terminal, build a corresponding dissemination tree for each of one or more second terminals within a sensing range; and transmit the corresponding dissemination tree to each of the one or more second terminals within the sensing range.

Abstract: Provided is a method of allocating a stack in a multi-threaded sensor operating system environment. The method dynamically allocates as much stack space as is required by a thread, thereby enabling efficient use of stack space. The method includes the steps of: whenever a function is called while a thread is performing an operation, dynamically allocating a stack space to be used by the called function; and when execution of the function is finished, returning the allocated stack space.