CN111897784B - Key value storage-oriented near data computing cluster system - Google Patents
Key value storage-oriented near data computing cluster system Download PDFInfo
- Publication number
- CN111897784B CN111897784B CN202010668559.7A CN202010668559A CN111897784B CN 111897784 B CN111897784 B CN 111897784B CN 202010668559 A CN202010668559 A CN 202010668559A CN 111897784 B CN111897784 B CN 111897784B
- Authority
- CN
- China
- Prior art keywords
- ndp
- key value
- file
- storage
- compression
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/10—File systems; File servers
- G06F16/17—Details of further file system functions
- G06F16/174—Redundancy elimination performed by the file system
- G06F16/1744—Redundancy elimination performed by the file system using compression, e.g. sparse files
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/30—Monitoring
- G06F11/3003—Monitoring arrangements specially adapted to the computing system or computing system component being monitored
- G06F11/3006—Monitoring arrangements specially adapted to the computing system or computing system component being monitored where the computing system is distributed, e.g. networked systems, clusters, multiprocessor systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/10—File systems; File servers
- G06F16/11—File system administration, e.g. details of archiving or snapshots
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/48—Program initiating; Program switching, e.g. by interrupt
- G06F9/4806—Task transfer initiation or dispatching
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Quality & Reliability (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides a near data computing cluster system for key value storage, which comprises: a host side and a plurality of NDP devices; the host end is respectively connected with each NDP device; the host end includes: the device comprises a cluster device management module, a file distribution module and a file migration module; the file transfer module is used for acquiring the compression requirements of each compression storage layer of each NDP device; and the file migration module is used for migrating the file to be compressed to the NDP equipment of the corresponding storage key value interval according to the key value range to perform storage, compression and sequencing. According to the invention, the calculable storage array is formed by arranging the plurality of NDP devices, so that the storage capacity and the calculation capacity of the whole system are ensured, meanwhile, the storage space of the host end is prevented from being occupied, and the CPU bottleneck of the host for the compression and sequencing operation of the database is greatly relieved, thereby being beneficial to improving the data processing efficiency of the whole system.
Description
Technical Field
The invention relates to the technical field, in particular to a key value storage-oriented near data computing cluster system.
Background
The rapid development of computer technology and the internet has prompted the emergence of semi-structured and unstructured data, and the proportion of the data in the total amount of data is increasing exponentially. However, under the condition of increasing unstructured data scale, the traditional relational database cannot meet the requirements of high-efficiency storage, high concurrency and high expandability of mass data. In contrast, key-value storage does not require a predefined data structure, and has been widely applied to unstructured data storage and management for providing low-latency read-write speed and supporting massive data traffic. At present, a key value storage system widely uses a log structure merge tree (LSM-tree) to realize data storage and management, and converts random writing into sequential writing, thereby obtaining excellent writing performance. For efficient management of data, a key-value storage system based on a log-structured merge tree generates a compress (compact) operation during operation to update and migrate a file table (SSTable) to a next level. However, the compression operation occupies a large amount of I/O bandwidth at the host end and the storage device end, which leads to performance degradation, and the update of the file table also causes the problem of enlarging the write data size.
Disclosure of Invention
Based on the technical problems in the background art, the invention provides a near data computing cluster system for key value storage.
The invention provides a near data computing cluster system for key value storage, which comprises: a host side and a plurality of NDP devices; the host end is respectively connected with each NDP device;
the host end includes:
the cluster equipment management module is used for generating an information object of the NDP equipment, and comprises the steps of setting a key value threshold of each NDP equipment and calculating a storage key value interval of each NDP equipment by combining the key value threshold; adding and deleting equipment; managing the working state of the equipment;
the file distribution module is used for receiving a file sent by an upper layer application and sending the file to the NDP equipment of the corresponding storage key value interval for storage according to the threshold range of the file;
the file migration module is used for acquiring compression requirements of compression storage layers of the NDP equipment, screening files to be compressed from the compression storage layers corresponding to the NDP equipment according to the compression requirements, and acquiring files overlapped with key value ranges of the files to be compressed from the next compression storage layer as supplementary files to be compressed;
the file migration module is used for migrating the file to be compressed and the supplementary file to be compressed to the NDP equipment in the corresponding storage key value interval according to the key value range to perform storage, compression and sequencing.
Preferably, the file migration module includes: the system comprises a storage monitoring unit, a file copying unit, a file dividing unit and a task sending unit;
the storage monitoring unit is used for acquiring the compression requirements of each compression storage layer of each NDP device and acquiring files to be compressed and compressed supplementary files corresponding to the compression requirements;
the file copying unit is connected with the storage monitoring unit and is used for screening files with key value ranges exceeding the storage key value interval of the currently located NDP equipment from the files to be compressed and the compressed supplementary files as files to be migrated and acquiring the files;
the file dividing unit is connected with the file copying unit, and is used for acquiring a file containing at least one key value threshold value in a key value range from the file to be migrated as a cutting target and dividing the cutting target according to the key value threshold value;
and the task sending unit is respectively connected with the file copying unit and the file dividing unit and is used for acquiring the file to be migrated and the divided file except for the cutting target and distributing the file to the NDP equipment corresponding to the storage key value interval according to the key value range.
Preferably, the task sending unit is further configured to divide all the files to be migrated into two parts and send the two parts to the compression storage layer at the host end and the NDP device corresponding to the key value range for compression, respectively, when the key value ranges of all the files to be migrated are located in the same storage key value interval.
Preferably, the task sending unit is further configured to obtain a file compressed by the host side and send the file to the NDP device corresponding to the key value range for storage.
Preferably, the host further includes a load balancing module, configured to monitor a task process of each compression storage layer of each NDP device, and count a process occupation time and an idle time of each NDP device in the same task process; the NDP equipment is also used for generating a key value adjusting instruction according to a comparison result of the process occupation time of each NDP equipment and a preset process time consumption threshold and a comparison result of the idle time of each NDP equipment and a preset idle threshold;
the cluster device management module is connected with the load balancing module and used for adjusting the key value threshold of the NDP device according to the key value adjusting instruction.
Preferably, the load balancing module is further connected to the file distribution module and the file migration module, respectively, and is configured to monitor and monitor current data deployment and data flow direction;
the load balancing module is used for counting key value sparse conditions and local key value threshold adjusting times of each NDP equipment compression task, and carrying out balancing processing on compression task distribution and data flow direction according to a counting result.
Preferably, the NDP device whose process occupation time is greater than the process time consumption threshold or whose idle time is greater than the idle threshold is used as the adjustment object, the NDP device whose storage key value range is adjacent to the adjustment object is used as the adjacent object, and the cluster device management module is configured to adjust the key value threshold between the adjustment object and any adjacent object according to the key value adjustment instruction.
Preferably, the key value adjusting instruction comprises an adjusting object and task time consumption of two adjacent objects in the task process; when the process occupation time of the adjustment object is larger than the process time consumption threshold, the cluster equipment management module adjusts the key value threshold between the adjustment object and an adjacent object with less task time consumption, and reduces the key value storage interval of the adjustment object; when the idle time of the adjustment object is larger than the idle threshold, the cluster equipment management module adjusts the key value threshold between the adjustment object and the adjacent object with the larger task time consumption, and enlarges the key value storage interval of the adjustment object.
Preferably, the host end and the NDP device both include a cache management module, the host end includes a cache management module for expanding the host cache function, and the NDP device end cache management module is configured to undertake caching of read data in a compression sorting operation.
Preferably, the host side communicates with the NDP device via an ethernet switch.
According to the near data computing cluster system for key value storage, the plurality of NDP devices are arranged to form the storage array, so that the storage capacity of the whole system is guaranteed, meanwhile, the occupied storage space of a host computer end is avoided, and the data processing efficiency of the whole system is improved.
In the invention, the files on the compression storage layer of each NDP device are managed through the file migration module, so that the ordered storage of the files on each NDP device is realized, the unified management efficiency of the files is improved, the files are conveniently and uniformly distributed, and the balance of compression tasks obtained by each NDP device is further ensured.
According to the invention, the cluster management module is used for monitoring the state of each NDP device and adjusting the threshold value of the key value, so that the flexible control of each NDP device is realized, the balance adjustment of the task process of each NDP device is ensured through the key value adjustment, and the minimum time consumption of a single compression task is ensured through the balance of the task process of each NDP device, so that the compression efficiency is improved, and the maximum utilization of the storage space of each NDP device is realized.
Drawings
Fig. 1 is a block diagram of a near data computing cluster system oriented to key value storage according to the present invention.
Detailed Description
Referring to fig. 1, a near data computing cluster system for key value storage according to the present invention includes: a host side and a plurality of NDP devices. The host end is respectively connected with each NDP device.
The host end includes:
the cluster equipment management module is used for generating an NDP equipment information object, and comprises the steps of setting key value thresholds of the NDP equipment and calculating storage key value intervals of the NDP equipment by combining the key value thresholds; adding and deleting equipment; and managing the working state of the equipment.
Specifically, in this embodiment, the NDP devices are sorted according to the key value threshold from small to large, the lower limit value of the key value interval of each NDP device is the key value threshold of the NDP device adjacent to the NDP device, and the upper limit value of the key value interval of each NDP device is the own key value threshold; the lower limit value of the key value storage interval of the NDP device corresponding to the minimum key value threshold is 0.
That is, it is assumed that the key value threshold of the NDP device 1 is f1, and the key value threshold of the NDP device n is fn; f1 < f2 < \\ 8230and f (n-1) < fn. Then, the NDP device 1 has a key value interval of (0, f 1), and the NDP device n has a key value interval of (f (n-1), fn ].
And the file distribution module is used for receiving the file sent by the upper layer application and sending the file to the NDP equipment of the corresponding storage key value interval for storage according to the threshold range of the file.
Specifically, suppose that the upper layer application newly sends a file A to the host, the key value range of the file A is [ a1, a2], and f (i-1) < a1 < a2 < fi; the file a is sent to the L0 compressed storage layer of NDP device i for storage.
Suppose that the upper layer application sends a new file B to the host end, the key value range of the file B is [ B1, B2], and f (i-2) < B1 < f (i-1) < B2 < fi. It can be seen that the key value range of the file B coincides with the storage key value intervals of the NDP device i-1 and the NDP device i, and at this time, the file B is sent to the L0 compressed storage layer of the NDP device i-1 and the L0 compressed storage layer of the NDP device i for storage.
And the file migration module is used for acquiring the compression requirements of each compression storage layer of each NDP device, screening the file to be compressed from the compression storage layer corresponding to each NDP device according to the compression requirements, and acquiring the file which is overlapped with the key value range of the file to be compressed from the next compression storage layer as the supplementary file to be compressed.
Specifically, in this embodiment, when the number of files stored in any one of the compression storage layers of any one of the NDP devices reaches the corresponding upper limit value, the compression requirement is generated. In this embodiment, the storage compression layer of each NDP device may be monitored in real time by the cluster device management module, and the compression requirement may be obtained. The file migration module obtains the compression requirements from the cluster management module. Therefore, the monitoring of the working state of the NDP equipment is separated from the file processing, and the file processing efficiency is further improved.
The file migration module is used for migrating the file to be compressed and the supplementary file to be compressed to the NDP equipment in the corresponding storage key value interval according to the key value range to perform storage, compression and sequencing.
Specifically, in the present embodiment, when the storage file of the LO compression storage layer reaches the upper limit value, the LO compression storage layer is directly compressed and stored in the L1 compression storage layer of the NDP device.
For the L1 compression storage layer and the compression storage layers above, files to be compressed and supplementary files to be compressed are firstly sorted according to the key value range. Assuming that a file on the Lj compression storage layer of a storage array composed of a plurality of NDP devices reaches an upper storage limit, a compression requirement is generated, a file C on the Lj compression storage layer of an NDP device i is selected as a file to be compressed, and a file D on the L (j + 1) compression storage layer of the NDP device i is selected as a supplemental file to be compressed.
The key value range of the file C is [ C1, C2], f (i-2) < C1 < C2 < f (i-1), the key value range of the file D is [ D1, D2], and f (i-2) < D1 < f (i-1) < D2 < fi.
It can be seen that the key value range of the file C is located in the key value range of the NDP device i-1, and the key value range of the file D is partitioned between the key value ranges of the NDP device i-1 and the NDP device i. Therefore, in this embodiment, the file migration module acquires the file C from the NDP device i, sends the file C to the Lj compression storage layer of the NDP device i-1, and performs compression on the file C by the Lj compression storage layer of the NDP device i-1 and stores the compressed file in the local L (j + 1) compression storage layer. Meanwhile, the file migration module acquires the file D from the NDP device i and divides the file D into: the file D1 with the key value range of [ D1, f (i-1) ] and the file D2 with the key value range of (f (i-1), D2], the file D1 and the file D2 are respectively sent to an Lj compression storage layer of the NDP device i-1 and an Lj compression storage layer of the NDP device i, the NDP device i-1 compresses the file D1 and stores the compressed file to an L (j + 1) compression storage layer of the NDP device i, and the NDP device i compresses the file D2 and stores the compressed file to the L (j + 1) compression storage layer of the NDP device i.
Specifically, in this embodiment, the file migration module includes: the system comprises a storage monitoring unit, a file copying unit, a file dividing unit and a task sending unit.
And the storage monitoring unit is used for acquiring the compression requirements of each compression storage layer of each NDP device and acquiring the files to be compressed and the compressed supplementary files corresponding to the compression requirements.
Specifically, in the system, each NDP device is monitored through a cluster device management module, the cluster device management module obtains a compression requirement of the NDP device at the first time, a file migration module obtains the compression requirement from the cluster device management module, obtains a corresponding compression storage layer according to the analysis of the compression requirement, and then performs file sorting processing.
In this embodiment, the storage monitoring unit communicates with the cluster device management module in real time to obtain the compression requirement.
And the file copying unit is connected with the storage monitoring unit and is used for screening files with key value ranges exceeding the storage key value interval of the current NDP equipment from the files to be compressed and the compressed supplementary files as files to be migrated and acquiring the files. In this embodiment, the file copying unit copies the file to be migrated, the cluster management module supervises the copied file of the NDP device, and when the file is copied and sent to the new NDP device, the cluster management module notifies the original NDP device to delete the copied file, so as to avoid redundant storage.
And the file segmentation unit is connected with the file copying unit, is used for acquiring a file containing at least one key value threshold value in a key value range from the file to be migrated as a cutting target, and is used for segmenting the cutting target according to the key value threshold value.
And the task sending unit is respectively connected with the file copying unit and the file dividing unit and is used for acquiring the file to be migrated and the divided file except for the cutting target and distributing the file to the NDP equipment corresponding to the storage key value interval according to the key value range.
For example, in the above embodiment, after the file copying unit copies the file C and the file D from the NDP device i, the file dividing unit divides the file D into: files D1 with key value range [ D1, f (i-1) ] and files D2 with key value range (f (i-1), D2 ].
The task transmission unit acquires the file C from the file copying unit and transmits the file C to the NDP device i-1. The task transmission unit acquires the file D1 and the file D2 from the file division unit and respectively transmits the files to the NDP device i-1 and the NDP device i.
In this embodiment, the task sending unit is further configured to, when the key value ranges of all the files to be migrated are located in the same storage key value interval, divide all the files to be migrated into two parts, and send the two parts to the compression storage layer at the host end and the NDP device corresponding to the key value range for compression. For example, taking file C as an example, after the task sending unit obtains file C, the task sending unit divides file C into file C1 and file C2, where file C1 is compressed by the host, and file C2 is sent to NDP device i-1 for compression and stored in the local L (j + 1) compression storage layer.
In this embodiment, the task sending unit is further configured to obtain a file compressed by the host and send the file to the NDP device corresponding to the key value range for storage. Thus, after the file C1 is compressed by the host side, the compressed file is sent to the NDP device i-1 through the task sending unit and is stored in the L (j + 1) compression storage layer by the NDP device i-1. Therefore, the host end and the NDP equipment are synchronously compressed, and the compression efficiency is improved. Meanwhile, the occupied storage space at the host end is also avoided.
In this embodiment, the host further includes a load balancing module, configured to monitor a task process of each compression storage layer of each NDP device, and count a process occupation time and an idle time of each NDP device in the same task process. The load balancing module is further configured to generate a key value adjustment instruction according to a comparison result between the process occupation time of each NDP device and a preset process time consumption threshold and a comparison result between the idle time of each NDP device and a preset idle threshold.
The cluster device management module is connected with the load balancing module and used for adjusting the key value threshold of the NDP device according to the key value adjusting instruction. Therefore, by adjusting the key value threshold, the file distribution module can distribute files according to the processing efficiency of different NDP devices, so that the task process of each NDP device is balanced.
Specifically, in this embodiment, the load balancing module is configured to count key value sparsity and local key value threshold tuning times of each NDP device compression task, and perform balancing processing on compression task distribution and data flow direction according to a statistical result.
Specifically, the load balancing module is configured to perform compression task distribution balancing by adjusting the priority of the compression queue, and is configured to calculate an optimal position of each key value threshold according to a ratio of the average file of the NDP device to the expected data migration to balance the data flow direction.
In this embodiment, the load balancing module is further connected to the file distribution module and the file migration module, respectively, and is configured to monitor and monitor current data deployment and data flow direction. Specifically, the load balancing module is connected with other modules to monitor data access optimization, data storage and the like in real time and control compression task distribution and data amount balance. The load balancing module dynamically adjusts the priority of the compressed task queue through judging the current data deployment and flow direction, and counts the adjustment times of the key value threshold value, thereby controlling the task selection work and the file distribution direction of the file migration module.
In this embodiment, when performing balanced distribution of compression tasks, the priority of the balanced compression queue is adjusted, and the tasks in the queue are sparsely compared according to the key values of the tasks and sorted, so as to determine the task to be completed by the balanced compression. When data flow equalization is executed, the optimal position of each key value threshold is determined by calculating the ratio of the average file of each NDP device to the expected data migration, then the priority of a compression queue is adjusted, and a task for adjusting the compression queue is selected and distributed to each device to be executed.
Thus, the load balancing module is used for balancing the key value sparsity ratio of each NDP device compression task; and analyzing the current data distribution and the data migration cost to calculate the optimal value of each key value threshold. In the working process, the load balancing module assists the system to change the flow direction of data by adjusting the key value threshold, control file migration and distribution and realize load balancing.
Specifically, in this embodiment, the NDP device whose process occupation time is greater than the process time consumption threshold or whose idle time is greater than the idle threshold is used as the adjustment object, the NDP device whose storage key value range is adjacent to the adjustment object is used as the adjacent object, and the cluster device management module is configured to adjust the key value threshold between the adjustment object and any adjacent object according to the key value adjustment instruction.
Thus, assuming that 5 NDP devices are provided in a system, and the process occupation time of the NDP device 1 is greater than the process consuming time threshold when a certain compression task is executed, the NDP device 1 is used as an adjustment object, the NDP device 2 is used as an adjacent object, and the key value threshold f1 of the NDP device 1 is adjusted to f1', and f1' < f1. In this way, the range of the storage key values of the NDP device 1 is narrowed to reduce files obtained by compressing the storage layer by the L0 of the NDP device 1, thereby reducing the amount of compression tasks; meanwhile, as the range of the storage key values is expanded, the NDP device 2 compresses the storage layer by using the L0 thereof to obtain more files so as to share the calculation pressure of the NDP device 1, thereby realizing the task equalization of the NDP device 1 and the NDP device 2.
Assuming that 5 NDP devices are provided in a system, and the idle time of the NDP device 5 is greater than the idle threshold when a certain compression task is executed, the NDP device 5 is used as an adjustment object, the NDP device 4 is used as an adjacent object, and the key value threshold f4 of the NDP device 4 is adjusted to f4', f4' > f4. Thus, the NDP device 4, due to the expansion of the range of the stored key values, has the L0 compression storage layer to obtain more files to share the computing stress of the NDP device 5. The range of the stored key values of the NDP device 5 is reduced to reduce the number of files obtained by compressing the storage layer by the L0 of the NDP device 5, thereby reducing the amount of compression tasks and realizing task equalization of the NDP device 4 and the NDP device 5.
Therefore, in the embodiment, by means of the balanced sharing of the tasks of the adjacent object and the adjusted object, compared with the unified adjustment of the key value thresholds of all the NDP devices, the file migration data volume in the subsequent tasks is reduced, and the calculation pressure caused by file migration is favorably reduced.
In this embodiment, the key value adjustment instruction includes an adjustment object and a task time consumption of two adjacent objects in the task process. When the process occupation time of the adjustment object is larger than the process time consumption threshold, the cluster equipment management module adjusts the key value threshold between the adjustment object and the adjacent object with less task time consumption, and the key value storage interval of the adjustment object is reduced. When the idle time of the adjustment object is larger than the idle threshold, the cluster equipment management module adjusts the key value threshold between the adjustment object and the adjacent object which consumes more time, and enlarges the key value storage interval of the adjustment object. Therefore, the key value equalization processing efficiency is further ensured.
Specifically, in this embodiment, the cluster device management module is configured to generate an information object of each NDP device. The information object is used for managing the number of the corresponding NDP device and the key value threshold, and calculating the key value storage interval of each NDP device by combining the key value threshold. The information object is also used for monitoring the communication between the corresponding NDP equipment and the host side, and controlling the file deployment and the file compression of the NDP equipment.
In this embodiment, the host side and the NDP device communicate with each other through an ethernet switch.
In this embodiment, both the host side and the NDP device include a cache management module. The host end comprises a cache management module for expanding the host cache function and bearing the cache work of the read operation data. And the NDP device end cache management module is used for bearing the cache of the read data in the compression sorting operation. Therefore, the read operation is ended at the host end as much as possible, and meanwhile, the NDP equipment end caches the data of the latest compression task, so that the spatial locality and the temporal locality of the compression sorting process are effectively utilized.
In this embodiment, the load balancing module is further configured to sense a behavior rule of the read operation, calculate an access frequency for target data of the read operation, and instruct the cache to perform hot and cold data layering.
The above description is only for the preferred embodiments of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be considered as the technical solutions and the inventive concepts of the present invention within the technical scope of the present invention.
Claims (7)
1. A key-value storage oriented near data computing cluster system, comprising: a host side and a plurality of NDP devices; the host end is respectively connected with each NDP device;
the host side includes:
the cluster equipment management module is used for generating an information object of the NDP equipment, and comprises the steps of setting a key value threshold of each NDP equipment and calculating a storage key value interval of each NDP equipment by combining the key value threshold; adding and deleting equipment; managing the working state of the equipment;
the file distribution module is used for receiving a file sent by an upper layer application and sending the file to the NDP equipment of the corresponding storage key value interval for storage according to the threshold range of the file;
the file migration module is used for acquiring compression requirements of compression storage layers of the NDP equipment, screening files to be compressed from the compression storage layers corresponding to the NDP equipment according to the compression requirements, and acquiring files overlapped with key value ranges of the files to be compressed from the next compression storage layer as supplementary files to be compressed;
the file migration module is used for migrating the file to be compressed and the supplementary file to be compressed to the NDP equipment in the corresponding storage key value interval according to the key value range for storage, compression and sequencing;
the host end also comprises a load balancing module which is used for monitoring the task process of each compression storage layer of each NDP device and counting the process occupation time and the idle time of each NDP device in the same task process; the NDP equipment is also used for generating a key value adjusting instruction according to a comparison result of the process occupation time of each NDP equipment and a preset process time consumption threshold and a comparison result of the idle time of each NDP equipment and a preset idle threshold;
the cluster equipment management module is connected with the load balancing module and used for adjusting the key value threshold of the NDP equipment according to the key value adjusting instruction;
the cluster device management module is used for adjusting the key value threshold between the adjustment object and any adjacent object according to the key value adjustment instruction;
the key value adjusting instruction comprises an adjusting object and the task time consumption of two adjacent objects in the task process; when the process occupation time of the adjustment object is larger than the process time consumption threshold, the cluster equipment management module adjusts a key value threshold between the adjustment object and an adjacent object with small task time consumption, and reduces a key value storage interval of the adjustment object; when the idle time of the adjustment object is larger than the idle threshold, the cluster equipment management module adjusts the key value threshold between the adjustment object and the adjacent object with the task time consumption being large, and enlarges the key value storage interval of the adjustment object.
2. The key-value-storage-oriented near data computing cluster system of claim 1, wherein the file migration module comprises: the system comprises a storage monitoring unit, a file copying unit, a file dividing unit and a task sending unit;
the storage monitoring unit is used for acquiring the compression requirements of each compression storage layer of each NDP device and acquiring files to be compressed and compressed supplementary files corresponding to the compression requirements;
the file copying unit is connected with the storage monitoring unit and is used for screening files with key value ranges exceeding the storage key value interval of the currently located NDP equipment from the files to be compressed and the compressed supplementary files as files to be migrated and acquiring the files;
the file dividing unit is connected with the file copying unit, and is used for acquiring a file containing at least one key value threshold value in a key value range from the file to be migrated as a cutting target and dividing the cutting target according to the key value threshold value;
and the task sending unit is respectively connected with the file copying unit and the file dividing unit and is used for acquiring files to be migrated and divided except the cutting target and distributing the files to the NDP equipment corresponding to the stored key value intervals according to the key value range.
3. The key-value-storage-oriented near data computing cluster system of claim 2, wherein the task sending unit is further configured to, when the key value ranges of all the files to be migrated are located in the same storage key value interval, divide all the files to be migrated into two parts, and send the two parts to the compressed storage layer at the host end and the NDP device corresponding to the key value range for compression, respectively.
4. The near data computing cluster system for key-value storage according to claim 3, wherein the task sending unit is further configured to obtain a host-side compressed file and send the host-side compressed file to the NDP device corresponding to the key-value range for storage.
5. The key-value-storage-oriented near data computing cluster system of claim 1, wherein the load balancing module is further connected to the file distribution module and the file migration module, respectively, for monitoring current data deployment and data flow direction;
the load balancing module is used for counting key value sparse conditions and local key value threshold adjusting times of each NDP equipment compression task, and carrying out balancing processing on compression task distribution and data flow direction according to a counting result.
6. The key-value-storage-oriented near data computing cluster system of claim 1, wherein the host side and the NDP device each comprise a cache management module, the host side comprises a cache management module for expanding a host cache function, and the NDP device side cache management module is configured to undertake caching of read data during a compression sorting operation.
7. The key-value-storage-oriented near data computing cluster system of any one of claims 1-6, wherein the host-side and the NDP device communicate via an Ethernet switch.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010668559.7A CN111897784B (en) | 2020-07-13 | 2020-07-13 | Key value storage-oriented near data computing cluster system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010668559.7A CN111897784B (en) | 2020-07-13 | 2020-07-13 | Key value storage-oriented near data computing cluster system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111897784A CN111897784A (en) | 2020-11-06 |
| CN111897784B true CN111897784B (en) | 2022-12-06 |
Family
ID=73192481
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010668559.7A Active CN111897784B (en) | 2020-07-13 | 2020-07-13 | Key value storage-oriented near data computing cluster system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111897784B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113779024B (en) * | 2021-08-05 | 2024-02-09 | 安徽大学 | Asynchronous parallel optimization method for key value storage system under near data processing architecture |
| CN118051643B (en) * | 2024-02-23 | 2024-11-05 | 中国科学院信息工程研究所 | Metadata sparse distribution-oriented LSM data organization method and device |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105159915A (en) * | 2015-07-16 | 2015-12-16 | 中国科学院计算技术研究所 | Dynamically adaptive LSM (Log-structured merge) tree combination method and system |
| CN107479833A (en) * | 2017-08-21 | 2017-12-15 | 中国人民解放军国防科技大学 | Key value storage-oriented remote nonvolatile memory access and management method |
| US9870168B1 (en) * | 2014-12-22 | 2018-01-16 | Emc Corporation | Key-value store with internal key-value storage interface |
| CN110928483A (en) * | 2018-09-19 | 2020-03-27 | 华为技术有限公司 | Data storage method, data acquisition method and equipment |
| CN110995871A (en) * | 2019-12-24 | 2020-04-10 | 浪潮云信息技术有限公司 | Method for realizing high availability of KV storage service |
| CN111400312A (en) * | 2020-02-25 | 2020-07-10 | 华南理工大学 | Edge storage database based on improved L SM tree |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7555493B2 (en) * | 2004-03-08 | 2009-06-30 | Transreplicator, Inc. | Apparatus, systems and methods for relational database replication and proprietary data transformation |
-
2020
- 2020-07-13 CN CN202010668559.7A patent/CN111897784B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9870168B1 (en) * | 2014-12-22 | 2018-01-16 | Emc Corporation | Key-value store with internal key-value storage interface |
| CN105159915A (en) * | 2015-07-16 | 2015-12-16 | 中国科学院计算技术研究所 | Dynamically adaptive LSM (Log-structured merge) tree combination method and system |
| CN107479833A (en) * | 2017-08-21 | 2017-12-15 | 中国人民解放军国防科技大学 | Key value storage-oriented remote nonvolatile memory access and management method |
| CN110928483A (en) * | 2018-09-19 | 2020-03-27 | 华为技术有限公司 | Data storage method, data acquisition method and equipment |
| CN110995871A (en) * | 2019-12-24 | 2020-04-10 | 浪潮云信息技术有限公司 | Method for realizing high availability of KV storage service |
| CN111400312A (en) * | 2020-02-25 | 2020-07-10 | 华南理工大学 | Edge storage database based on improved L SM tree |
Non-Patent Citations (4)
| Title |
|---|
| 《Co-KV: A Collaborative Key-Value Store Using Near-Data Processing to Improve Compaction for the LSM-tree》;HuiSun等;《arXiv》;20180711;第1-8页 * |
| 《DStore: A Holistic Key-Value Store Exploring Near-Data Processing and On-Demand Scheduling for Compaction Optimization》;Hui Sun 等;《IEEE》;20181004;第1-21页 * |
| 《基于近数据计算的LSM-tree键值存储系统Compaction优化方法》;刘伟;《中国优秀硕士学位论文全文数据库 (信息科技辑)》;20190715;第I137-39页 * |
| 基于SSD-SMR混合存储的LSM树键值存储系统的性能优化;王洋洋等;《计算机科学》;20180715(第07期);第61-65+89页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111897784A (en) | 2020-11-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11340812B2 (en) | Efficient modification of storage system metadata | |
| US10374792B1 (en) | Layout-independent cryptographic stamp of a distributed dataset | |
| KR102034833B1 (en) | Apparatus for Accessing Data Using Internal Parallelism of Flash Storage based on Key-Value and Method thereof | |
| CN109947668B (en) | Method and device for storing data | |
| US10019459B1 (en) | Distributed deduplication in a distributed system of hybrid storage and compute nodes | |
| US11429630B2 (en) | Tiered storage for data processing | |
| JP2021501389A (en) | Data hierarchy storage and hierarchy search method and device | |
| CN111897784B (en) | Key value storage-oriented near data computing cluster system | |
| US11816029B2 (en) | Adjustment of garbage collection parameters in a storage system | |
| EP4209922A1 (en) | Columnar caching in tiered storage | |
| CN113486026A (en) | Data processing method, device, equipment and medium | |
| CN112559459B (en) | Cloud computing-based self-adaptive storage layering system and method | |
| CN105450784A (en) | An apparatus and method for distributing consumption nodes to messages in an MQ | |
| US11221991B2 (en) | Techniques for selectively activating and deactivating entropy computation | |
| CN115129621A (en) | Memory management method, device, medium and memory management module | |
| WO2024169851A1 (en) | Data compression method, system, and device, and computer readable storage medium | |
| CN116431615A (en) | Flexible data partition routing method for complex service scene | |
| CN111949218B (en) | Load balancing method, system, medium and terminal for high-availability storage system | |
| CN118394784A (en) | Data query method, system, device cluster, medium and program product | |
| CN111897783B (en) | KV storage system optimization method for executing parallel tasks based on multi-NDP collaborative host | |
| CN118502683B (en) | Task processing method and system for memory chip | |
| US20240086091A1 (en) | Using object properties for facilitating object storage | |
| Yan et al. | Optimizing the restoration performance of deduplication systems through an energy-saving data layout | |
| US11262935B2 (en) | Optimized distributed deduplication for distributed cluster | |
| KR102024846B1 (en) | File system program and method for controlling data cener using it |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |