Abstract: The described technology is generally directed towards accelerating data handling in a cloud data storage system by using smart network interface cards (SmartNICs) at the nodes. Instead of copying data to kernel space, many input/output (I/O) operations can be handled primarily by the SmartNIC, using the SmartNIC's memory. For example, mirrored data writes can be sent directly from the SmartNIC's memory associated with the node handling the write to other nodes, without first copying the data to kernel space. Object reads can be handled at a node by having segments of the object queued, in order, in the node's associated SmartNIC's memory, and sent to a requesting client without having to be copied to the handling node's kernel space, unless low memory conditions exist in the SmartNIC's memory.

Abstract: Facilitating transmission of multi-segment data portions for storage devices is provided herein. A system can comprise a processor and a memory that stores executable instructions that, when executed by the processor, facilitate performance of operations. The operations can comprise sending a read request for object data associated with an object, wherein the object data comprises multiple segments distributed across nodes of a storage cluster. The operations also can comprise receiving an out of order transmission that comprises the multiple segments. Further, the operations can comprise ordering segments of the multiple segments based on respective packet sequence numbers associated with the segments, resulting in a properly ordered object data. According to some implementations, the respective packet sequence numbers can be respective precalculated ranges of packet sequence numbers associated with the segments of the multiple segments.

Abstract: Facilitating transmission of multi-segment data portions for storage devices is provided herein. A system can comprise a processor and a memory that stores executable instructions that, when executed by the processor, facilitate performance of operations. The operations can comprise sending a read request for object data associated with an object, wherein the object data comprises multiple segments distributed across nodes of a storage cluster. The operations also can comprise receiving an out of order transmission that comprises the multiple segments. Further, the operations can comprise ordering segments of the multiple segments based on respective packet sequence numbers associated with the segments, resulting in a properly ordered object data. According to some implementations, the respective packet sequence numbers can be respective precalculated ranges of packet sequence numbers associated with the segments of the multiple segments.

Abstract: The described technology is generally directed towards accelerating data handling in a cloud data storage system by using smart network interface cards (SmartNICs) at the nodes. Instead of copying data to kernel space, many input/output (I/O) operations can be handled primarily by the SmartNIC, using the SmartNIC's memory. For example, mirrored data writes can be sent directly from the SmartNIC's memory associated with the node handling the write to other nodes, without first copying the data to kernel space. Object reads can be handled at a node by having segments of the object queued, in order, in the node's associated SmartNIC's memory, and sent to a requesting client without having to be copied to the handling node's kernel space, unless low memory conditions exist in the SmartNIC's memory.

Abstract: The described technology is generally directed towards accelerating data handling in a cloud data storage system by using smart network interface cards (SmartNICs) at the nodes. Instead of copying data to kernel space, many input/output (I/O) operations can be handled primarily by the SmartNIC, using the SmartNIC's memory. For example, mirrored data writes can be sent directly from the SmartNIC's memory associated with the node handling the write to other nodes, without first copying the data to kernel space. Object reads can be handled at a node by having segments of the object queued, in order, in the node's associated SmartNIC's memory, and sent to a requesting client without having to be copied to the handling node's kernel space, unless low memory conditions exist in the SmartNIC's memory.