34 results about "Remote memory" patented technology

The invention discloses a user-noninductive cold and hot sensing remote memory exchange system, which belongs to the field of computer storage, and is characterized in that cluster nodes are divided into extensible remote memory service nodes and computing client nodes according to roles; the remote memory node comprises a remote memory exchange server and a memory resource registration module; the computing client node comprises a FrontSwap-based remote memory exchange client, and a remote memory exchange server side is registered as a memory resource which is non-inductive to a user; furthermore, popularity statistics based on a popularity histogram is realized in a kernel mode; the memory exchange behavior is guided through the dynamic decision-making module based on online reinforcement learning, the memory management efficiency and the system performance are remarkably improved, and efficient and non-inductive memory expansion service is provided for user application.

Embodiments of the present disclosure provide a data transmission method, a device, a storage medium, and a program product. The embodiments of the present disclosure provide a software-implemented RDMA memory page fault solution: when acquiring data from a remote memory, a remote end maps a page-faulted virtual memory into a specific physical memory region, allows for the occurrence of a memory page fault in an RDMA process, and adds a memory page fault processing flow between a local end and the remote end to implement memory page fault processing; and when writing data into the remote memory, the remote end provides a shared virtual memory and a private virtual memory and exposes the shared virtual memory to the local end, the local end writes data into the shared virtual memory, and then a remote control object migrates the data to the private virtual memory, wherein memory page faults are allowed to occur and be processed during the migration. Since memory page faults are allowed to occur and page fault processing is supported, dynamic allocation of physical memory resources is supported during the RDMA process, thereby improving the utilization of memory resources while leveraging the advantages of RDMA.

The embodiment of the invention discloses a memory access optimization method and device, electronic equipment, a storage medium and a program.The method comprises the steps that memory access data of an application program are dynamically monitored in real time, and remote memory access data of the application program are determined according to the memory access data of the application program; under the condition of determining that the remote memory access data of the application program meets the memory exchange condition, determining a target exchange local memory from local memories; determining a target exchange remote memory according to the remote memory access data of the application program; and performing memory exchange processing on the target exchange local memory and the target exchange remote memory. According to the technical scheme provided by the embodiment of the invention, efficient management of the memory can be realized in a high-memory-demand use scene, and memory access delay is reduced, so that the memory access efficiency is improved, and the overall performance of the system is improved.

The application discloses an inter-node communication method and an electronic device, relates to the technical field of inter-node communication, and comprises the following steps: when a first node is initialized, a first completion queue group and a first queue pair group are created to form a first resource pool, and a first memory pool is applied as a unified storage area; when the first node joins a cluster, a node announcement message is sent to other nodes in the cluster; in response to the first node receiving a communication task pointing to a second node, a first queue pair is selected from available queue pairs, and a first memory area is divided from the first memory pool; the second queue pair number and the second remote access key of the second node are obtained from a preset connection table; and based on the first queue pair and the first memory area, a remote memory direct access operation is initiated to access the opposite end identified by the second queue pair number with the second remote access key, thereby solving the technical problem of long inter-node communication delay in the prior art, and improving the efficiency and reliability of inter-node communication.

A method of data storage is disclosed comprising: adding a hash to a Merkle mountain range (MMR) stored in a local memory, wherein the MMR is persisted to a second MMR stored in a remote memory, the remote memory also comprising cryptographic information; copying the second MMR and the cryptographic information to the local memory and obtaining a version identifier identifying the copied data; determining a total storage size of the data added to the MMR; determining whether a portion of the remote memory has capacity for the data; and if the portion has capacity: writing the second MMR with the hash back to the remote memory and providing the version identifier; or if the portion has insufficient capacity: writing the hash, the cryptographic information, and further cryptographic information to a new portion of the remote memory; and transmitting the version identifier. A corresponding device is also disclosed.

This invention discloses a method and system for exchanging local and remote memory objects in a single-copy mode. First, the invention constructs a logical address space and deploys it on memory nodes, establishing a unified correspondence between spatial addresses and virtual memory addresses. A memory management unit then translates virtual memory addresses into physical addresses. Second, a unique-copy mode is used to trigger object swapping operations, and an asynchronous garbage collection mechanism balances memory release efficiency with operational stability. Finally, a dedicated memory allocator improves memory management efficiency, including a split memory layout, a kernel offloading mechanism, and multi-partition load balancing. This invention avoids inter-process address switching overhead through a unified address space; eliminates data redundancy and multi-copy synchronization overhead through a unique-copy mode; and improves memory allocation efficiency through a split memory layout, kernel offloading, and multi-partition management mechanisms, achieving efficient utilization of remote memory resources and high-performance scheduling in multi-process shared scenarios.

Systems and methods for updating remote memory side caches in a multi-GPU configuration are disclosed herein. In one embodiment, a graphics processor for a multi-tile architecture includes a first graphics processing unit (GPU) having a first memory, a first memory side cache memory, a first communication fabric, and a first memory management unit (MMU). The graphics processor includes a second graphics processing unit (GPU) having a second memory, a second memory side cache memory, a second memory management unit (MMU), and a second communication fabric that is communicatively coupled to the first communication fabric. The first MMU is configured to control memory requests for the first memory, to update content in the first memory, to update content in the first memory side cache memory, and to determine whether to update the content in the second memory side cache memory.

The invention provides a memory scheduling method and device, electronic equipment, a medium and a program product, and relates to the technical field of computers. The memory scheduling method comprises the steps of dividing memory hierarchies for the hybrid memory system on the basis of a performance index of a far-end memory; dynamically scanning the container based on the attribute of the memory data in the container to obtain a container access heat map; configuring a cold and hot threshold value of the container for the importance weight allocated to the container and the memory hierarchy based on the container access heat map; and performing memory page scheduling on the container based on the cold and hot threshold and the container access heat map. According to the technical scheme, scheduling of the memory page in the container gives consideration to data heat, container importance and dynamic data features, and rationality of cold and hot scheduling of the memory page is improved while global scheduling of the memory page is achieved.

The present application provides a method for constructing a remote memory access protection mechanism, a remote memory node, and a device. The method includes: obtaining a permission application from a computing node, wherein the permission application is generated based on the computing node submitting an access request from a user program to a network card queue pair, and the access request includes a remote memory address; if the permission application is authenticated, the memory window is bound to the remote memory address, and a permission symbol is returned, which is used for remote memory access; if the remote memory access ends, the memory window is unbound from the remote memory address to invalidate the permission symbol. In this way, the permission mechanism can be used to effectively isolate illegal remote access from user programs, and the permission symbol becomes invalid after the access ends, effectively avoiding potential problems such as remote data corruption and remote data privacy leakage faced by the remote memory architecture.

The invention relates to the technical field of equipment management service, in particular to an enterprise-level equipment as a service (DaaS) system which comprises a remote memory monitoring module, an equipment use identification module, a use cycle mapping module, an abnormal state identification module and a staged service label module. According to the invention, by collecting the memory usage data of the device in real time and calculating the memory utilization rate, the resource abnormity can be identified in time, and by analyzing the starting and ending time of the device session and combining the reference of the service level agreement for comparison, the refined classification of the device usage condition is realized; through deviation comparison with baseline data such as equipment use frequency and task throughput, abnormal behaviors are found in advance, abnormal behavior recognition is performed through multi-dimensional feature cross validation, the accuracy of abnormal detection can be greatly improved, the accuracy, response speed and early warning capability of equipment resource management are remarkably improved, the operation and maintenance cost is effectively reduced, and the method is suitable for popularization and application. And the service life cycle of equipment is prolonged.

The invention relates to a separate memory architecture-oriented swap space cache optimization system and method, and the method comprises the steps: receiving a page swap request sent by a host memory space, and converting the page swap request into a swap space memory access request; a cardinal number tree cache index is generated according to the logic block number in the exchange space memory access request, searching is conducted according to the index, and if searching succeeds, a data block is obtained from a cache space and sent to a host memory space; if the search is not successful, adopting an asynchronous active cache replacement strategy to select a data block from the original cache region, writing the data block back to the remote memory pool, and meanwhile, adopting a dynamic pre-fetching strategy based on historical access to read pre-fetching data from the remote memory, and storing the pre-fetching data into a pre-fetching cache region. On the premise that an existing switching mechanism of an operating system is not changed, the access efficiency of the switching space of the separated memory is improved, the access frequency of the remote memory is reduced, and therefore the memory access time delay of the whole system is reduced, and the I / O performance of the system is improved.

The application relates to the technical field of distributed computing and memory management, and discloses a memory dynamic resource allocation and scheduling method for big data system integration, which comprises the following steps: analyzing a distributed computing job to establish a mapping of a data block logical address to a local page table entry; generating an event notification when a downstream operator completes reading of the data block; writing data in a local physical page frame into a registered memory area of a remote computing node through RDMA, and constructing a remote memory index table in a kernel state; writing a remote redirection mark and an integer index, and releasing the corresponding local physical page frame; when a target page table entry triggers a page fault exception and contains the redirection mark, the system acquires a remote data address according to the integer index by querying the index table, pulls the remote data to update the page table entry, and restores process execution. The application effectively releases local memory resources, realizes on-demand recovery of remote data through a page fault exception mechanism of an operating system, and reduces system recalculation overhead.

Unused memory registrations in a remote memory system are proactively identified and freed. A memory server maintains a list of resource tracking entries for its outstanding memory registrations. The memory server continually monitors its list of resource tracking entities. When the memory server finds an unused memory registration in its list, it unregisters the corresponding shared memory region. The monitoring may be performed in the background of the memory server by, e.g., a monitor thread. As a result, unused memory registrations may be quickly released.

The invention relates to the technical field of edge computing architectures, in particular to an edge computing method and system capable of achieving direct exchange of remote memory data, in an FPGA of a heterogeneous processing unit, a hardware RoCE v2 RDMA engine, a high-speed memory pool, a reconfigurable calculator, a flow control and congestion manager and other key components are creatively integrated, and by means of a QDMA IP core, the RDMA engine, the high-speed memory pool, the reconfigurable calculator, the flow control and congestion manager and other key components are integrated. A high-speed PCIe data channel is constructed with an edge reasoning module containing a CPU, a GPU and an AI accelerator, the defects that a traditional RDMA scheme is low in system integration degree and heavy in hardware dependence are overcome, efficient cooperation of calculation acceleration and the RDMA function is achieved, meanwhile, the problem that on-chip resources are limited in a large-scale connection scene can be solved, stable and efficient data transmission is guaranteed, preprocessing is supported during data flowing, and the data flow speed is increased. The method is deeply fused with AI reasoning, a calculation and communication separation pattern is broken, and the flow control and scheduling capability is strong.

The present invention relates to the technical field of communications. Disclosed is a smart network interface controller unloading-based remote memory system, comprising a computing node, a smart network interface controller and a memory node. The computing node sends a remote memory access request to the smart network interface controller; the smart network interface controller is used for analyzing the remote memory access request and accessing the memory node; the memory node is used for initializing and storing a remote memory. The smart network interface controller comprises a remote memory request processing module, a data address management module and an RDMA communication frontend; the memory node comprises an RDMA communication backend; the use stage of the remote memory system comprises an initialization stage, an application registration stage and an operation stage. The present invention unloads the management unit overhead originally on the memory node to the smart network interface controller, thus achieving in-network management of the remote memory; in addition, the memory node is only responsible for providing the memory, and is completely unaware of remote memory access by applications, thereby effectively reducing the computational management overhead of a CPU on the memory node.

The present disclosure relates to methods and systems for shared memory access APIs in distributed systems. A system allocates, in a distributed system comprising a plurality of nodes, a plurality of memory portions comprising shared remote memory content. The system registers the allocated portions with an operating system for access via RDMA. The system accesses, by a first node, the allocated portions to obtain a local copy. The system performs, via a library structured atomic application programming interface call, an atomic operation on one or more bits of the shared remote memory content by one or more of: updating the one or more bits based on a new value and an offset; fetching, from the shared remote memory content, a current value of the one or more bits prior to the update based on the offset; and performing an action on the shared remote memory content based on a comparison of the fetched current value to an expected value in the local copy.

Unused memory registrations in a remote memory system are proactively identified and freed. A memory server maintains a list of resource tracking entries for its outstanding memory registrations. The memory server continually monitors its list of resource tracking entities. When the memory server finds an unused memory registration in its list, it unregisters the corresponding shared memory region. The monitoring may be performed in the background of the memory server by, e.g., a monitor thread. As a result, unused memory registrations may be quickly released.

The embodiment of the present specification provides a memory allocation method and device, the method comprises: in the case of determining to perform page table page allocation for a target page table of a target process, acquiring a resource binding mechanism corresponding to the target process; according to a memory node binding constraint in the resource binding mechanism, determining a target memory node corresponding to the target process, and allocating a page table page for the target page table in the target memory node; in the case that local memory of the target memory node is insufficient to cause the page table page allocation to fail, migrating a target physical page in the local memory to a first remote memory node, and continuing to perform the step of allocating a page table page for the target page table in the target memory node until the page table page allocation succeeds. By such a forced allocation of a page table page in a target memory node and in combination with an intelligent memory migration mechanism, performance loss caused by remote access of a page table is fundamentally eliminated.

The invention belongs to the related technical field of key value storage in a separated memory environment, and particularly relates to a key value storage method based on a key value separation log structure merge tree in a separated memory, which comprises the following steps: organizing value data into a value data table capable of being remotely and directly accessed and operated by the memory, and adopting a fine-grained garbage collection mode to store the value data into a value data table; the positions of all key value pair data are not moved, and a remote memory space where each piece of failure value data is located is recognized and recycled through compression operation on the SST; designing a space, and writing the value data into the recovered fragmented memory space by using a writing strategy to improve the space utilization rate and avoid the serialization process of the value data, and constructing an invalid data space collector to manage and record a value data table written in the MN end; a fragment space reorganizer is used for collecting fragment space of a value data table and reorganizing the fragment space into a plurality of sets to receive value data of a CN end, and the fragment space of the value data table is effectively utilized through splitting and flashing operation.

A system includes a first network device and a second network device. The first network device is configured to send a command over a network that (i) specifies a key for accessing a hash table in a memory and (ii) instructs that a value be read or written to a location in the hash table corresponding to the key. The second network device is configured to receive the command over the network and execute the command by calculating the location in the hash table based on the key and reading or writing the value at the location.

The present invention relates to the field of large model parameter processing and discloses a terminal large model parameter collaborative loading method, scheduling method and system, comprising: determining the file size according to the model file to be loaded, and opening a memory area in the inference device based on the file size, and byte-aligning the starting address of the memory area; obtaining the cache status of the model to be loaded in the memory, and dividing the model to be loaded into a cached area and an uncached area according to the cache status; obtaining the hard disk loading speed of the inference device and the network status of the remote memory device, and dividing the uncached area into a hard disk loading area and a remote memory loading area; loading the model parameters of the cached area from the memory of the inference device to the corresponding position of the memory area, loading the model parameters of the hard disk loading area from the hard disk of the inference device to the corresponding position of the memory area, and loading the model parameters of the remote memory loading area from the remote memory device to the corresponding position of the memory area.

A system with an interface for remote memory. In some embodiments, the system includes: an interface circuit having: a first interface, configured to be connected to a processing circuit; and a second interface, configured to be connected to memory, the first interface including a cache coherent interface, and the second interface being different from the first interface.

The embodiment of the present application provides a data recovery method, a computing device and a storage medium. In the embodiment of the present application, data is written to a local memory node, record information for recording a write operation corresponding to the written data is generated and stored according to the write operation; the written data in the local memory node is written to a remote memory node; first positioning information is generated and stored, and the first positioning information is used for positioning the written data that has been written to the remote memory node; in the case that the computing node is abnormal, the written data in the local memory node that has not been written to the remote memory node is recovered according to the first positioning information and the record information. Therefore, the written data in the remote memory node is retained, and the life cycle of the remote memory node is maintained. The life cycle of the remote memory node is effectively utilized, and the data recovery amount is reduced.

The invention relates to memory registration for remote direct memory access. Unused memory registration in a remote memory system is actively identified and de-registered. A memory server maintains a list of resource tracking entries for its uncompleted memory registration. A memory server continuously monitors a list of its resource tracking entries. When the memory server finds an unused memory registration in its list, it cancels the corresponding shared memory area. The monitoring may be performed in the background of the memory server by, for example, a monitoring thread. Thus, unused memory registrations may be quickly released.

Unused memory registrations in a remote memory system are proactively identified and freed. A memory server maintains a list of resource tracking entries for its outstanding memory registrations. The memory server continually monitors its list of resource tracking entities. When the memory server finds an unused memory registration in its list, it unregisters the corresponding shared memory region. The monitoring may be performed in the background of the memory server by, e.g., a monitor thread. As a result, unused memory registrations may be quickly released.

A present invention embodiment provides trace-base analysis and memory migration. A trace is performed on a process that is executing one or more preconfigured tasks to collect trace data comprising memory access addresses of the process and function calls of the process. The trace data is analyzed to identify one or more metadata memory regions based on frequent memory accesses and a content memory region based on infrequent memory accesses during the trace. It is determined that a portion of content addresses of the content memory region can be migrated to a remote memory without disrupting execution of the process. The portion of content addresses is migrated to the remote memory. The process is patched to insert remote logic that redirects access operations to the portion of the content addresses to the migrated portion of content addresses in the remote memory. The process is executed.

The present invention discloses a separated memory system and a management method. The system includes a host side, a DPU, and a remote memory side interconnected through RDMA. The implementation steps of the method are as follows: The cache management module initializes the DRAM memory space; the cache management module manages the read / write requests issued by the computing server; the remote memory mapping management module maintains the mapping table and transfers the read / write requests; the remote memory service module processes the read / write requests; the remote memory recycling module releases the invalid memory pages in the memory server. In the present invention, the DRAM serves as the local cache of the remote memory, which can achieve persistent caching of the data frequently accessed by the computing server, enabling the high-frequency access requests to be directly responded through the DRAM, reducing the RDMA network transmission, and thus reducing the read / write latency. At the same time, the CPU aggregates the available memory of the remote memory side through the RDMA network interface, expanding the available memory of the operating system of the computing server.

A memory device, an electronic device, and a near memory device are provided. The memory device includes a near memory device and a far memory device, where the far memory device stores a compressed embedded vector, the near memory device includes a buffer that receives a request for pooling the embedded vector and outputs a hit signal or a missing signal in response to the request, the hit signal comprises an embedded vector corresponding to the request; an address calculation circuit that calculates a starting memory address and a memory size of the embedded vector in response to the missing signal; a decoding circuit that obtains a compressed embedded vector corresponding to the request from the remote memory device based on the starting memory address and the memory size, and outputs an embedded vector corresponding to the request by decoding the compressed embedded vector; and a pooling circuit that outputs a pooled embedded vector by performing a pooling operation on the embedded vector.

A memory access method and device are provided. A memory access method may include: identifying, when an access to a page of a remote memory occurs, a type of the access; allocating a sparse buffer when the access is a sparse write; storing data for the sparse write in the sparse buffer; storing an address for the sparse write as a key and the sparse buffer as a value in a buffer table; and updating an instruction pointer to point to a next instruction.

The invention discloses a separated soft memory management method under a multi-application deployment scene. According to the method, the difference of a soft memory system and a separated memory system in system architecture is fully considered, a unified object metadata format is designed to be compatible with reconstruction and remote access semantics, and a metadata and actual data separated storage form is adopted to support efficient unloading and recycling of a remote memory; during running, the soft memory of the application program is managed by utilizing the object pool, memory fragments are reduced by adopting a log structured distribution mode, and higher distribution efficiency is realized; the memory recovery strategy is adaptively adjusted according to the medium type and the utilization rate of the soft memory, the strategy based on swap-out, evacuation and forced recovery is realized by setting different threshold values, and the memory allocation is dynamically adjusted by monitoring local and remote soft memory resources and combining the potential revenue of an application program. And efficient utilization of soft memory resources in a multi-application deployment environment is ensured.