In a solution for accessing a storage system, a client sends, based on an obtained start address that is of a queue of an NVMe storage device and to which an access request points and an obtained logical address that is of the NVMe storage device and to which the access request points, a remote direct memory access command to a storage node in which the NVMe storage device is located.

A garbage collection method is provided and applied to a data storage device. The garbage collection method includes the following steps: selecting source blocks from data blocks, wherein a total number of valid data of the source blocks is larger than or equal to a predetermined data number of a block; copying valid data of a part of the source blocks into a destination block, wherein a total number of the valid data of the part of the source blocks is smaller than the predetermined data number; copying all or a part of valid data of remaining source blocks into the destination block; updating a logical to physical addresses mapping table based on a mapping information of the destination block; and recovering all or a part of the source blocks as spare blocks.

Techniques for caching may include: determining an update to a first data page of a first cache on a first node, wherein a second node includes a second cache and wherein the second cache includes a copy of the first data page; determining, in accordance with one or more criteria, whether to send the update from the first node to the second node; responsive to determining, in accordance with the one or more criteria, to send the update, sending the update from the first node to the second node; and responsive to determining not to send the update, sending an invalidate request from the first node to the second node, wherein the invalidate request instructs the second node to invalidate the copy of the first data page stored in the second cache of the second node.

A data merge method for a rewritable non-volatile memory module including a plurality of physical units is provided. The method includes: selecting at least one first physical unit and at least one second physical unit from the physical units; reading first mapping information from the rewritable non-volatile memory module, and the first mapping information includes mapping information of the first physical unit and mapping information of the second physical unit; copying valid data collected from the first physical unit and valid data collected from the second physical unit to at least one third physical unit of the physical units according to the first mapping information; and when a data volume of valid data copied from the second physical unit to the third physical unit reaches a data volume threshold, stopping collecting valid data from the second physical unit, and continuing collecting valid data from the first physical unit.

Systems and methods for storing nodes, preferably, leaf nodes, of a data tree structure into storage are disclosed, and in one or more aspects restoring the leaf nodes from storage, preferably to memory. Copying the nodes into storage includes in an embodiment share-latching a first node of a data tree to be copied; copying the first node that is share-latched into storage; determining if there is a sibling second node linked to the first node; following a link between the first copied node and the sibling second node, share-latching the sibling second node, unlatching the first copied node, and copying the sibling second node into storage. Restoring includes copying the leaf nodes from storage, updating the leaf nodes, and creating/recreating the data tree.

A computer system includes a first group of CPU modules operatively coupled to at least one first Programmable ASIC Node Controller configured to execute transactions directly or through a first interconnect switch to at least one second Programmable ASIC Node Controller connected to a second group of CPU modules running a single instance of an operating system.

In an example, there is disclosed a host-fabric interface (HFI), including: an interconnect interface to communicatively couple the HFI to an interconnect; a network interface to communicatively couple the HFI to a network; network interface logic to provide communication between the interconnect and the network; a coprocessor configured to provide an offloaded function for the network; a memory; and a caching agent configured to: designate a region of the memory as a shared memory between the HFI and a core communicatively coupled to the HFI via the interconnect; receive a memory operation directed to the shared memory; and issue a memory instruction to the memory according to the memory operation.

Disclosed in some examples are methods, systems, devices, and machine-readable mediums that provide for techniques for scrambling and/or updating meta-data that enable an efficient internal copyback operation. In some examples, in order to update the meta-data, the meta-data and host-data are separated and the only the meta-data is sent to the controller to be updated during a modified internal copyback operation. The host-data is not transmitted to the controller. While sending the meta-data utilizes resources of the communication link between the memory dies and the controller, it uses much fewer resources than if the host-data were also transmitted.

A method and a system detects a cache line as a potential or confirmed hot cache line based on receiving an intervention of a processor associated with a fetch of the cache line. The method and system include suppressing an action of operations associated with the hot cache line. A related method and system detect an intervention and, in response, communicates an intervention notification to another processor. An alternative method and system detect a hot data object associated with an intervention event of an application. The method and system can suppress actions of operations associated with the hot data object. An alternative method and system can detect and communicate an intervention associated with a data object.

In exemplary aspects of managing the ejection of entries of a coherence directory cache, the directory cache includes directory cache entries that can store copies of respective directory entries from a coherency directory. Each of the directory cache entries is configured to include state and ownership information of respective memory blocks. Information is stored, which indicates if memory blocks are in an active state within a memory region of a memory. A request is received and includes a memory address of a first memory block. Based on the memory address in the request, a cache hit in the directory cache is detected. The request is determined to be a request to change the state of the first memory block to an invalid state. The ejection of a directory cache entry corresponding to the first memory block is managed based on ejection policy rules.