318 results about "Virtual memory" patented technology

The invention provides a complex scene-oriented AI large model lightweight deployment method, and relates to the technical field of edge computing, and the method comprises the steps: carrying out the structured pruning of a pre-trained Transform network based on the attention head importance score, carrying out the dynamic sparsification of the activation state of a feedforward network according to the input tensor entropy value, employing the dynamic mixing precision quantization, and carrying out the reconstruction of an AI large model. Obtaining network parameters after pruning quantization; deploying the pruned and quantized network parameters to an edge computing device, distributing a feature extraction operator to a neural network processor through a heterogeneous computing scheduler, and unloading a classification operator to a multi-core central processing unit; and managing an on-chip memory in combination with a virtual memory paging mechanism, realizing zero-copy data transmission by utilizing a direct memory access controller, and outputting a reasoning result tensor. According to the method, efficient and reliable operation of the large model at the resource-constrained edge node is realized.

A key-value cache management system processes inference requests of a generative model. An inference request requests a starting virtual memory address and a number of layers in the generative model. In response to receiving an inference request, contiguous virtual memory space is reserved for the key-value cache in accordance with the inference request by assigning the starting virtual memory address in the key-value cache and calculating memory pointers for each layer in the model and each block so that the one or more blocks are written sequentially. The generated memory pointers are outputted to the generative model so that each self-attention layer writes computed key-value pairs at physical addresses in the key-value cache specified by the memory pointers.

The present application belongs to the technical field of data processing and relates to a network message processing method and apparatus, and a computer device and a storage medium. The method comprises: on the basis of a received target network message, acquiring a target memory block from a pre-constructed shared memory pool, so as to generate a message memory; on the basis of a network-interface-card driver, performing first identification processing on the target network message, so as to obtain a target network message descriptor; on the basis of the target network message descriptor, performing second identification processing on the message memory, so as to obtain a first socket buffer; and performing parsing processing on the first socket buffer by means of a network protocol stack, and on the basis of a parsing processing result, reading network message data from a virtual memory corresponding to the message memory, so as to complete the processing of the network message. In the present application, on the basis of a constructed shared memory pool, zero-copy network packet reception can be realized during network message processing, thereby avoiding performance loss caused by the memory copy of messages.

The invention discloses a data processing method, device and equipment and a readable storage medium, and the method comprises the steps: running a business application in a virtual machine, and obtaining a business application instruction set; the instruction type of the business application instruction set is an instruction type supported by a processor of the host machine; writing a service application instruction set in a virtual memory of the virtual machine into a shared physical memory in a virtual interaction device corresponding to the virtual machine, and generating writing notification information used for representing a writing completion state of the service application instruction set; the shared physical memory is a physical memory used for being jointly read and written by the virtual machine and the management program in the host machine; and in the management program, obtaining write-in notification information, obtaining the business application instruction set in the shared physical memory through the write-in notification information, and executing the business application instruction set through a processor of the host machine to obtain instruction execution data corresponding to the business application instruction set. By adopting the method and the device, the performance of running the business application by the virtual machine in the host machine can be improved.

Systems and methods herein are for at least one circuit to perform at least one process of different processes that may be associated with one or more applications, where the process may use one part of a virtual memory based on different requests by the different processes, where the virtual memory may include parts of equal allocations based on a maximum of different memory sizes in specifications associated with the different processes, where the one part of the virtual memory may be in a mapping with respect to one part of a physical memory of different allocated sizes, and where a translation for the mapping can occur using a start address and the maximum of the different memory sizes.

The invention relates to a memory management method and device applied to a server-free data center, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: in response to a virtual memory application instruction, performing virtual memory allocation in a local computing resource pool to obtain a virtual address space; in response to the virtual address access instruction, acquiring cache information of the computing resource pool, and triggering missing page interruption under the condition that data corresponding to the virtual address is not hit in the cache information; the cache information comprises locally cached cache data and page table information; acquiring physical memory resources in a memory resource pool according to the virtual address and the page table information under the condition of triggering missing page interruption of the anonymous page; and updating the page table information and the cache information according to the physical memory resources. By adopting the method, the processing delay of the data center can be reduced while the decoupled memory management logic is used for improving the resource utilization rate.

The invention provides a hot upgrade method for a virtual machine monitor (VMM), which comprises the following steps of: pausing the operation of a virtual machine, storing state information of an original VMM, and reserving a kernel-based virtual machine KVM extension page table and a memory management structure of the original VMM; reserving a page global directory and a virtual memory area VMA corresponding to a virtual machine RAM area according to the memory management structure, and refreshing a process address space described by the memory management structure into a new VMM execution program so as to complete switching of the original VMM into the new VMM; and executing the new VMM, recovering the state information to the new VMM, and inheriting the KVM extension page table by the new VMM to recover the operation of the virtual machine. The invention further provides computer equipment, a computer readable medium and a computer program product.

The invention discloses a data processing method and a related device, and the method comprises the steps: firstly obtaining a target instruction of a virtual machine, and enabling the target instruction to be used for controlling the virtual machine; and then according to the target virtual memory address corresponding to the target instruction, an encryption key corresponding to the target instruction is determined, and different virtual memory addresses correspond to different encryption keys. And according to the encryption key, performing instruction coding on the target instruction through a coding algorithm to obtain a coding result, and storing the coding result as the target instruction in the target virtual memory address. Finally, in the running process of the virtual machine, when a calling request for the target instruction is obtained, the coding result is read from the target virtual memory address, instruction decoding is conducted on the coding result through a decoding algorithm corresponding to the coding algorithm according to the encryption key, and the target instruction is obtained. The encryption key is generated based on the storage address of the instruction, and the encryption key is used for instruction coding to realize the specificity of a coding result, so that the security of the virtual machine is improved.

Implementations of this specification provide key-value cache management, model reasoning, and data processing methods and apparatuses for large language models. In an implementation, a method comprises allocating a virtual memory block in a virtual address slot to newly-added token key-value data of a model reasoning request, in response to determining that a scheduling result of the model reasoning request indicates the model reasoning request is scheduled for execution, maintaining a mapping relationship between an occupied virtual address slot and a physical graphics memory block allocated to the model reasoning request, and copying the newly-added token key-value data to the physical graphics memory block.

The invention discloses a memory access method and device, a storage medium and a program product, and relates to the technical field of memory access, comprising: monitoring memory access information of a target memory; the memory access information comprises a missing page address sequence during memory access, an address distribution characteristic when an address conversion failure event occurs in an address conversion lookaside buffer, a cross-node access event and a virtual memory access frequency; determining corresponding target feature information; the target feature information comprises spatial locality, thermal density, access dispersion and unbalance degree data; and determining a current memory access mode based on the target feature information, and determining a target access strategy from preset memory access strategies configured with different memory page table prefetching rules to access the memory. The memory access feature information is quantified through the memory access information, the corresponding memory access mode is determined, and the corresponding memory page table prefetching rule is executed, so that the traversal delay of the multi-level page table can be reduced, and the memory access performance is improved by fully utilizing the memory bandwidth.

The invention discloses a server system based on a public cloud technology and an access method thereof, and belongs to the technical field of cloud services. The server system comprises a server and an external device inserted in the server. A virtual machine manager in the server is used for providing a virtual external device and a virtual memory for a virtual machine in the server, the virtual external device is obtained through device simulation based on the external device, and the virtual memory is obtained through device simulation based on a memory configured for the external device. The virtual device driver of the virtual external device is used for sending an access request carrying the target GPA to the external device when the virtual external device accesses the target GPA of the virtual memory. And the external device is used for obtaining a target host physical address corresponding to the target client physical address based on the access request, obtaining target data recorded by the target host physical address, and sending the target data to the virtual device driver. According to the invention, the address translation performance of the server system is effectively improved.

The invention discloses a data transmission system, method and device, a storage medium and a computer program product, and relates to the technical field of communication. The data transmission system comprises a virtual machine, a virtual RDMA device and a host, wherein the virtual RDMA device runs between the virtual machine and the host. Wherein the virtual machine is used for running an application; the virtual RDMA equipment is used for registering a virtual memory and a virtual RDMA connection corresponding to the virtual machine, configuring a memory mapping relation between the virtual memory and a physical memory and a connection mapping relation between the virtual RDMA connection and the physical RDMA connection, and issuing the memory mapping relation and the connection mapping relation to the host; and the host is used for transmitting the RDMA data of the application according to the memory mapping relation and the connection mapping relation. Decoupling of the front-end virtual machine and the rear-end host is realized, the expansibility of connection is improved, a data transmission system is more modularized and is easy to manage, and the overhead of the host is reduced.

The invention relates to a progressive memory delay monitoring and diagnosis method oriented to a virtual memory subsystem. The method comprises the following steps: mounting a BPF program to a plurality of key event points on a kernel; when the key event point is triggered, acquiring state information of a corresponding event; packaging the state information into an event, and writing the event into a perf ring buffer associated with the outputmap; and the user mode program asynchronously reads the packaged event from the perf ring buffer, performs classification and aggregation according to the process and the event type, generates a delay terminal log and a histogram, and outputs a diagnosis result. Through systematic architecture design, the state keeping capability and the kernel context access capability of the eBPF are combined with an event model of a virtual memory subsystem, and a set of closed-loop monitoring system oriented to a diagnosis target is constructed. High-precision and high-reliability memory operation delay measurement is realized.

A file system in a user space partition of virtual memory may be mounted by a computing device that runs a virtual machine which includes a set of storage disks. The file system in user space may then expose one or more virtual files associated with one or more storage disks that correspond to one or more loop devices configured to map files of the virtual machine to the one or more virtual files. The computing device may then receive a request to read a data block stored at the virtual machine and may identify a file and corresponding virtual file that stores the requested data block based on a set of metadata provided by the loop devices. The computing device may then determine the location of the data block stored at the virtual machine, and may read the data block from the determined location.

Examples are disclosed herein relating to memory paging. In some examples, a host device is configured to communicate with an expansion device. The expansion device can include first and second memory, and a device virtual memory address (DVA) table. The expansion device can store data that can be requested by the host device. The first memory is a cache for the second memory based on a page presence table (PPT). The PPT can indicate a presence of second memory pages in the first memory cache. The DVA table can include information to locate data in the first memory based on a host physical memory address of a memory request. The device physical memory address can identify a memory location at which the data is stored. The data can be provided from the expansion device to the host device in response to the memory request based on the PPT.

The embodiment of the invention provides a core dump file generation method and device, electronic equipment and a medium, and relates to the technical field of data processing.The method comprises the steps that in response to abnormity of a target process, a thread causing the abnormity of the target process is determined as a target thread from threads of the target process; retrieving a virtual memory address from a register occupied by the target thread and a stack memory area of the target thread; determining a target address of a data page which belongs to a heap memory area and has mapping from the retrieved virtual memory address; and generating a core dump file containing the data page mapped by the target address. By applying the scheme provided by the embodiment of the invention, the validity of the core dump file can be improved while the size of the core dump file is reduced.

The embodiment of the invention relates to the technical field of terminal security management, and discloses a kernel management method and device based on Hypervisor, and the method comprises the steps: obtaining abnormal event information of an operating system kernel of terminal equipment; determining virtual address field information corresponding to the abnormal event according to the abnormal event information; performing virtual memory measurement on the operating system kernel according to the virtual address field information to obtain a memory measurement result; and determining whether data in the kernel of the operating system is tampered or not according to the memory measurement result. According to the embodiment of the invention, effective detection and protection of data tampering in the kernel of the operating system are realized, and the overall security and system integrity of the terminal equipment are improved.

The invention provides a memory error simulation method and device based on a satellite-borne intelligent computing system, a medium and a simulator, and relates to the technical field of satellites. The method comprises the following steps: inputting a deep learning neural network engine into a deep learning reasoning program of a satellite-borne intelligent computing system; executing a deep learning reasoning program to read a process page table; based on the DRAM standard file, the DRAM mapping file and the address mapping relation, determining a memory active area of the deep learning neural network engine in the virtual memory model; performing construction processing on the memory active area, constructing an error injection space represented in a bitmap tree form, and determining an error information injection position in the error injection space according to particle flipping error model information and a Monte Carlo random algorithm; and performing reverse translation processing on the physical address in the virtual memory model, and determining a target virtual address of the deep learning reasoning process. And a single-particle upset memory error and a multi-particle upset memory error caused by space radiation can be effectively simulated.

The invention discloses a protocol conversion system which comprises a first virtual machine, a second virtual machine and a virtual machine manager, a second virtual machine communication driver, a protocol conversion module and a front-end driver are arranged in the second virtual machine, and an application program is installed in the second virtual machine and sends data to the front-end driver during operation. The front-end driver communicates with the protocol conversion module, the protocol conversion module sends data to the second inter-virtual-machine communication driver, the second inter-virtual-machine communication driver transmits the processed data to the virtual machine manager, and the processed data is transmitted to the first virtual machine after being processed by the virtual machine manager. The first virtual machine processes the data and then sends the data to the physical peripheral in the virtual machine manager, and the protocol conversion module is added between the front-end driver and the virtual memory of the virtual machine, so that data communication can be realized without resetting or writing a driver program, and the system development difficulty and maintenance cost are reduced.

Examples of the present disclosure disclose a memory system and an operation method thereof and a storage device and an operation method thereof. The memory system includes a memory device and a memory controller, the kth memory blocks in memory planes in the memory device form one super memory block, super memory blocks form one virtual memory block, the mth physical pages coupled with the nth word lines in the memory planes from a same super memory block in the virtual memory block form one virtual physical page, the virtual physical pages from super memory blocks in a same virtual memory block form one strip; and the memory controller is configured to: receive memory data that needs to be written to a strip; generate parity check data of the strip according to the memory data; and send the memory data and the parity check data to the memory device.

The embodiment of the invention provides an address mapping method and device, electronic equipment and a readable storage medium, and the method comprises the steps: determining a target state tag corresponding to a physical address of a to-be-accessed client based on a state tag region in a virtual memory space of a host machine under the condition of binary translation of a source access instruction of the client; under the condition that the target state mark is matched with a target memory access operation indicated by the source memory access instruction, determining a host machine virtual mapping address corresponding to the to-be-accessed client physical address in a host machine virtual memory space based on a direct address mapping rule; and executing a target memory access operation on a target host machine physical address corresponding to the host machine virtual mapping address. Therefore, the host machine virtual mapping address corresponding to the physical address of the to-be-accessed client is directly determined through the direct address mapping rule, retrieval traversal is not needed, the overhead in the address mapping process can be reduced, the address conversion efficiency is improved, and meanwhile the number of times of interaction between the client and the host machine is reduced.

The invention provides a dynamic adaptive memory and hard disk hybrid writing method and device. The method comprises the following steps: generating a spatio-temporal feature vector based on a memory access sequence; obtaining a popularity calculation formula according to the spatial-temporal feature vector, and calculating a data popularity value of the data memory by using the popularity calculation formula; obtaining a data popularity probability according to the data popularity value of the data memory; and dividing the memory data into four levels of superheat, heat, temperature and cold according to the data popularity probability, obtaining a comprehensive utilization rate by integrating three factors of a physical memory utilization rate, a system idle bandwidth and queue response delay, and executing a corresponding migration strategy according to the comprehensive utilization rate. Through real-time data popularity analysis, a difference bit pre-migration strategy and a dynamic priority adjustment algorithm, efficient identification and scheduling of cold and hot data of the virtual memory are realized, and the hybrid writing performance between a physical memory and a hard disk is optimized.

Certain aspects of the present disclosure provide techniques and apparatus for translation lookaside buffer (TLB) compression. Embodiments include determining that a plurality of physical memory addresses, associated with a plurality of virtual memory addresses, are contiguous with one another and share one or more common address bits or one or more common attribute bits, wherein each respective physical memory address of the plurality of physical memory addresses corresponds to a separate respective physical memory page. Embodiments include generating a tag for an entry in a TLB, the tag representing the plurality of virtual memory addresses. Embodiments include associating, in the entry in the TLB, the tag with data comprising: a single instance of the one or more common address bits or the one or more common attribute bits of the plurality of physical memory addresses; and other bits of the plurality of physical memory addresses.

The embodiment of the invention provides a memory adjustment method, a database memory adjustment method and a memory adjustment device.The memory adjustment method comprises the steps that in response to a memory adjustment request, a target virtual address of a to-be-released memory is determined; traversing a pre-configured virtual memory area corresponding to at least one process, and determining a target virtual address range to which the target virtual address belongs, the virtual memory area including a plurality of pre-divided virtual address ranges; the shared physical memory corresponding to the target virtual address range is released, and the shared physical memory comprises the memory to be released. The shared physical memory corresponding to the whole virtual memory area does not need to be released, small-granularity memory release is achieved, memory adjustment delay is reduced, memory adjustment efficiency is improved, meanwhile, the small-granularity memory release avoids the problem that memory adjustment needs waiting, memory adjustment complexity is reduced, and memory adjustment efficiency is improved. And the memory adjustment efficiency is improved.

There is provided a method, system and non-transitory storage medium for serving a large language model (LLM) engine by allocating memory on demand. A virtual memory space with virtual pages is initialized for a tensor. A first request with first input tokens is received, added to the tensor by being allocated to first virtual pages and mapped to first page frames in physical memory. The tensor is provided to the LLM engine for processing. A second request with second input tokens is received, and a first output token is received from the LLM engine. The first output token is added to the tensor by being allocated to a second virtual page contiguous to the first virtual pages and mapped to a second page frame. The second input tokens are allocated to second virtual pages and mapped to second page frames in physical memory. The tensor is provided for further processing.

Embodiments of this application disclose a method and an apparatus for restoring a running status of an application, and a storage medium, and relate to the field of software technologies. In embodiments of this application, in a running process of an application, management information of a first virtual memory area corresponding to a first physical memory area is stored. When the application exits abnormally, a physical page of the first physical memory area is maintained, to store the physical page and prevent the physical page from being reclaimed. When the application is restored and started, a mapping relationship between the first virtual memory area and the first physical memory area can be re-established based on the management information of the first virtual memory area.

The embodiment of the invention provides a memory management method, a host machine, electronic equipment, a storage medium and a program product, and relates to the technical field of computers.The method comprises the steps that in response to a memory capacity adjusting request of a virtual machine, a target physical memory block is determined from a physical address space of the host machine, the memory capacity adjustment request is triggered in the running process of the virtual machine and is used for requesting to adjust the capacity of a first memory area of the virtual machine, and the first memory area is used for allocating a virtual memory for kernel mode data of the virtual machine; and based on the physical address of the target physical memory block, adjusting the address mapping relationship of the first memory area on the physical address space of the host machine. According to the technical scheme provided by the embodiment of the invention, the kernel mode data is isolated in the first memory area, and the capacity of the first memory area is dynamically adjusted in the running process of the virtual machine, so that the elastic capacity expansion and shrinkage of the kernel mode memory are realized.

To provide an information processing device that enables overlapping of communication and calculation without buffering the communication between calculation processes on a host.SOLUTION: In an information processing device, a coprocessor-host memory translation function maps a calculation process virtual memory for a calculation process in a coprocessor provided in the information processing device to an offloading function virtual address for collective communication offloading. The collective communication offloading function uses the mapped offloading function virtual address to perform remote direct memory access (RDMA) for another information processing device connected via a network, and implements communication between the calculation process and a calculation process of the other information processing device.SELECTED DRAWING: Figure 1

The embodiment of the invention provides a memory pooling management system, a memory pooling management method, electronic equipment and a storage medium. The memory pooling management system comprises a first server used for creating an expansion device memory through a first operating system and configuring a first address mapping relation between a first virtual memory and the expansion device memory; the second server is used for dividing a pooling host memory in a host memory of the second server through a second operating system; and the expansion switch is in communication connection with the first server through a first bus, is in communication connection with the second server through a second bus, and is used for configuring a second address mapping relation between the expansion equipment memory and the pooling host memory. The first server searches the first address mapping relation based on the virtual address to obtain an intermediate physical address, the extension switch routes the intermediate physical address to the access physical address, and the second server accesses the pooling host memory based on the access physical address.

The invention discloses an automatic AI memory leak intelligent positioning and root cause traceability analysis system, and relates to the technical field of memory leak intelligent positioning and root cause traceability analysis. The automatic AI memory leak intelligent positioning and root cause traceability analysis method includes the following steps: S1, data acquisition: performing index acquisition through a multi-dimensional data acquisition unit, and supporting physical memory and virtual memory utilization rate, heap and stack memory allocation and log release, object creation, track destruction and thread execution state index acquisition; according to the automatic AI memory leak intelligent positioning and root cause tracing analysis system, the leak position is positioned, the root cause can be analyzed from dimensions such as code defects, improper configuration and resource competition, the leak recurrence rate is reduced, normal memory fluctuation and a real leak mode are effectively distinguished through sliding window time sequence analysis, exception screening and feature engineering optimization, and the reliability of the system is improved. False alarms are reduced, and the accuracy of abnormal event identification is improved.