Data transformation offloading in an artificial intelligence infrastructure that includes one or more storage systems and one or more graphical processing unit (GPU) servers, including: storing, within the storage system, a dataset; identifying, in dependence upon one or more machine learning models to be executed on the GPU servers, one or more transformations to apply to the dataset; and generating, by the storage system in dependence upon the one or more transformations, a transformed dataset.

Data transformation caching in an artificial intelligence infrastructure that includes one or more storage systems and one or more graphical processing unit (GPU) servers, including: identifying, in dependence upon one or more machine learning models to be executed on the GPU servers, one or more transformations to apply to a dataset; generating, in dependence upon the one or more transformations, a transformed dataset; storing, within one or more of the storage systems, the transformed dataset; receiving a plurality of requests to transmit the transformed dataset to one or more of the GPU servers; and responsive to each request, transmitting, from the one or more storage systems to the one or more GPU servers without re-performing the one or more transformations on the dataset, the transformed dataset.

A drive-to-drive data transfer may be performed by means of a host configuring registers in a source device and a sink device and triggering a transfer. Data copy logic of controllers of the source and sink devices may then proceed to perform the endpoint to endpoint transfer, for example, using transaction layer packets (TLPs).

Disk based emulation of tape libraries is provided with features that allow easier management and administration of a backup system and also allow increased flexibility to both archive data on tape at a remote location and also have fast restore access to archived data files. Features include automatic emulation of physical libraries, and the retention and write protection of virtual tapes that correspond to exported physical tapes.

Devices, computer-readable media, and methods for reducing the number of hops that internal messages must traverse in data center switching architectures are disclosed. In one example, a data center includes a first rack housing a first server, a first computational process associated to a first storage drive hosted on the first server and residing within a first level of the distributed storage system, a second rack housing a second server, a second computational process associated to a second storage drive hosted on the second server and residing within the first level of the distributed storage system, and a first switch communicatively coupled to the first level to receive messages directly from the first computational process and the second computational process.

An artificial intelligence and machine learning infrastructure system, including: one or more storage systems comprising, respectively, one or more storage devices; and one or more graphical processing units, wherein the graphical processing units are configured to communicate with the one or more storage systems over a communication fabric; where the one or more storage systems, the one or more graphical processing units, and the communication fabric are implemented within a single chassis.

The present disclosure includes apparatuses, methods, and systems for data relocation in memory. An embodiment includes a controller, and a memory having a plurality of physical units of memory cells. Each of the physical units has a different sequential physical address associated therewith, a first number of the physical units have data stored therein, a second number of the physical units do not have data stored therein, and the physical address associated with each respective one of the second number of physical units is a different consecutive physical address in the sequence. The controller can relocate the data stored in the physical unit of the first number of physical units, whose physical address in the sequence is immediately before the first of the consecutive physical addresses associated with the second number of physical units, to the last of the consecutive physical addresses associated with the second number of physical units.

A data management circuit with network functions and a network-based data management method are provided. The network-based data management method is employed to manage a storage device coupled to a computer that includes a processor. The method includes steps of: receiving a network packet through a network; sending the network packet to the processor or accessing the storage device, according to a network header of the network packet; and requesting the processor to access the storage device according to a remaining capacity of the storage device and/or a content of the network packet.

Embodiments of the present disclosure relate to managing volatile and non-volatile memory. A set of volatile memory sensor data may be obtained. A set of non-volatile memory sensor data may be obtained. The set of volatile memory sensor data and the set of non-volatile memory sensor data may be analyzed. A memory condition may be determined to exist based on the analysis. In response to determining that the memory condition exists, one or more memory actions may be issued.

A technique involves: in response to receiving a first request for adjusting a first width of a disk array to a second width, obtaining, based on source identification information of a source stripe group in the disk array in the first request, source block identification information of a source block associated with the source stripe group. The technique further involves: determining destination identification information of a destination stripe group associated with the second width for storing data. The technique further involves: storing, based on the source identification information and the destination identification information, source data and metadata for the source data from the source block into a destination block of the destination stripe group, the metadata including node identification information for accessing nodes of the source block. The technique further involves: adjusting the node to access the destination block based on the node identification information.