A QoS directory flow control method and control device in a distributed file system

Abstract

The present invention discloses a method and device for controlling QoS directory traffic in a distributed file system. The method comprises: receiving a feedback message reported by an access node, the feedback message including read / write information of the access node reading and writing a target QoS directory; updating a mapping entry of the access node in a feedback message mapping table of the target QoS directory based on the feedback message, the mapping entry being used to characterize the mapping relationship between the node identifier of the access node and the read / write information; generating QoS configuration information for each access node corresponding to the target QoS directory based on each mapping entry in the feedback message mapping table, and sending the generated QoS configuration information to the corresponding access node to distribute the traffic of the target QoS directory to each access node. The technical solution provided by the present invention can improve the efficiency of traffic distribution.

Description

Technical Field

[0001] The present invention relates to the field of distributed file systems, and in particular to a QoS directory flow control method and control device in a distributed file system. Background Art

[0002] In cloud scenarios, multiple users often share cloud resources. Distributed file storage resources are a key category of cloud resources. For distributed file storage resources, file storage capacity is typically allocated at the directory level based on user purchases. More importantly, different traffic levels are allocated, including bandwidth and Input / Output Operations Per Second (IOPS). Distributed file storage resources in cloud scenarios are supported by a distributed file system. Typically, administrators assign different directories to different users within the distributed file system. These directories are logically isolated and interconnected, without nesting. However, the disk space, CPU resources, and network resources of the cluster nodes supporting the file storage resources are still shared. To ensure that each user receives the planned quality of service, Quality of Service (QoS) is required to limit the bandwidth and IOPS resources used by users.

[0003] For distributed file storage, bandwidth and IOPS limits need to be implemented at the directory level. For distributed file storage clusters with multiple access nodes, it's necessary to balance the read and write traffic within each access node's QoS directory. Specifically, each access node's QoS allocation (i.e., the specific bandwidth and IOPS allocation for the access node) is adjusted accordingly based on the access node's read and write I / O usage. This avoids wasting the QoS configuration value (i.e., the overall bandwidth and IOPS cap for the QoS directory) and ensures that the bandwidth and IOPS resources purchased by users are fully utilized. Existing traffic distribution methods are typically implemented using a token bucket. A token bucket is an internal storage pool within a network device, and tokens are virtual packets that fill the token bucket at a given rate. A token bucket can be thought of as a container for tokens with a pre-set capacity. The system adds tokens to the bucket at a set rate. When the bucket is full, excess tokens are released. When data arrives at the device, a number of tokens corresponding to the data size are first removed from the token bucket to transmit the data. This means that sufficient tokens must be present in the token bucket for data transmission. If the number of tokens is insufficient, the data will be discarded or cached, thereby limiting the message flow to a rate less than or equal to the rate at which tokens are generated, thereby achieving the purpose of limiting flow.

[0004] However, the token bucket method relies on a QoS central node in the distributed file system. When the QoS central node allocates tokens to QoS access nodes, if the number of cached tokens is too small, the QoS access nodes will frequently request tokens, significantly impacting the system's ability to process read and write requests. If the cached tokens are too large, changes to QoS configuration values ​​will take effect slowly, preventing timely adjustment of allocated values. Therefore, a more efficient method for controlling QoS directory traffic is urgently needed. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a method and apparatus for controlling QoS directory traffic in a distributed file system, which can improve the distribution efficiency of QoS directory traffic.

[0006] According to the first aspect, the present invention provides a QoS directory traffic control method in a distributed file system, which is applied to a metadata service node, and the method includes: receiving a feedback message reported by an access node, the feedback message including read and write information of the access node on the target QoS directory; based on the feedback message, updating the mapping entry of the access node in the feedback message mapping table of the target QoS directory, the mapping entry is used to characterize the mapping relationship between the node identifier of the access node and the read and write information; according to each mapping entry in the feedback message mapping table, generating QoS configuration information for each access node corresponding to the target QoS directory, and sending the generated QoS configuration information to the corresponding access node to distribute the traffic of the target QoS directory to each access node.

[0007] Optionally, based on the feedback message, updating the mapping entry of the access node in the feedback message mapping table of the target QoS directory includes: if the feedback message reported by the access node indicates that the access node has not performed any read or write behavior within a preset number of periods, removing the mapping entry corresponding to the access node from the feedback message mapping table; if the feedback message reported by the access node indicates that the access node has performed read or write behavior for the first time, adding a new mapping entry to the feedback message mapping table based on the feedback message; wherein, at the end of each period, the accumulated traffic value of the access node corresponding to each mapping entry is counted, and each mapping entry in the feedback message mapping table is modified based on the counted accumulated traffic value.

[0008] Optionally, according to each mapping entry in the feedback message mapping table, QoS configuration information is generated for each access node corresponding to the target QoS directory, including: if the current cycle is the initial cycle or the cycle for the addition of new nodes, the QoS configuration value is evenly distributed to the access nodes corresponding to each mapping entry, and the distribution result is used as the QoS configuration information corresponding to each access node; if the current cycle is not the initial cycle or the cycle for the addition of new nodes, the saturation state of the access node corresponding to each mapping entry is determined; the excess allocation amount of each unsaturated access node is recovered, and the excess allocation amount is reallocated among each saturated access node to generate the QoS configuration information of each access node in the next cycle.

[0009] Optionally, determining the saturation state of the access node corresponding to each mapping entry includes: calculating the traffic value of each access node in the previous period based on the traffic accumulation value to obtain multiple first traffic values; comparing the first traffic value with the QoS allocation value of the previous period, if the first traffic value is less than the QoS allocation value of the previous period, determining that the access node is an unsaturated access node; if the first traffic value is equal to the QoS allocation value of the previous period, determining that the access node is a saturated access node.

[0010] Optionally, the recycling of excess allocation amounts of each unsaturated access node and redistribution of the excess allocation amounts among each saturated access node includes: calculating an allocation increment, where the allocation increment is the ratio of the sum of the excess allocation amounts of all unsaturated access nodes to the number of saturated access nodes; adding the QoS allocation value of the saturated access node in the previous period to the allocation increment to generate a new QoS allocation value for each saturated access node in the next period; and using the traffic value of each unsaturated access node in the previous period as the new QoS allocation value for each unsaturated access node in the next period.

[0011] Optionally, the method of determining the saturation state of the access node corresponding to each mapping entry, recovering the excess allocation amount of each unsaturated access node and redistributing the excess allocation amount among each saturated access node includes: predicting the QoS allocation value of each access node in the next cycle based on the changing trend of the accumulated traffic value of each access node in multiple cycles to obtain multiple traffic prediction values; determining the saturated access nodes and unsaturated access nodes among the access nodes based on the relationship between the traffic prediction value and the QoS allocation value of the previous cycle; calculating the ratio of the traffic prediction value of each saturated access node to the total traffic prediction value of the saturated access node to obtain a split ratio; splitting all excess allocation amounts into multiple allocation increments based on the split ratio, and allocating them to the corresponding saturated access nodes, the QoS allocation value of the saturated access node in the next cycle being the sum of the QoS allocation value of the previous cycle and the allocation increment; and using the predicted traffic value of the unsaturated access node as the QoS allocation value of the unsaturated access node in the next cycle.

[0012] Optionally, determining the saturated access nodes and unsaturated access nodes among the access nodes based on the relationship between the traffic prediction value and the QoS allocation value of the previous cycle includes: comparing the traffic prediction value of the access node with the QoS allocation value of the previous cycle, if the traffic prediction value is less than the QoS allocation value of the previous cycle, determining that the access node is an unsaturated access node; if the traffic prediction value is equal to the QoS allocation value of the previous cycle, determining that the access node is a first saturated access node; if the traffic prediction value is greater than the QoS allocation value of the previous cycle, determining that the access node is a second saturated access node; for the first saturated access node, increasing its traffic prediction value to a second traffic prediction value according to a preset ratio.

[0013] Optionally, the method also includes: when the QoS configuration value is lowered, simultaneously lowering the QoS allocation value of each access node in the next cycle according to the downward adjustment ratio of the QoS configuration value; when the Qos configuration value is increased, calculating the ratio of the QoS allocation value of each saturated access node in the next cycle to the total QoS allocation value of the saturated access node, and generating a second split ratio; based on the second split ratio, allocating the increase in the Qos configuration value to the corresponding saturated access node.

[0014] Optionally, the generated QoS configuration information is sent down to the corresponding access node, including: based on the QoS configuration information, updating the second mapping entry of the access node in the traffic allocation mapping table of the target QoS directory, the second mapping entry being used to characterize the mapping relationship between the node identifier of the access node and the QoS configuration information; generating a traffic allocation message corresponding to the changed mapping entry in the traffic allocation mapping table, and sending the traffic allocation message down to the corresponding access node.

[0015] According to the second aspect, the present invention provides a QoS directory traffic control device in a distributed file system, and the device is applied to a metadata service node, and the device includes: an information receiving module, used to receive feedback messages reported by access nodes, and the feedback messages include read and write information of the access nodes reading and writing the target QoS directory; a feedback message mapping module, used to update the mapping entry of the access node in the feedback message mapping table of the target QoS directory based on the feedback message, and the mapping entry is used to characterize the mapping relationship between the node identifier of the access node and the read and write information; a traffic configuration module, used to generate QoS configuration information for each access node corresponding to the target QoS directory according to each mapping entry in the feedback message mapping table, and send the generated QoS configuration information to the corresponding access node to distribute the traffic of the target QoS directory to each access node.

[0016] According to the third aspect, an embodiment of the present invention provides a QoS directory traffic control device in a distributed file system, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the method described in the first aspect or any optional embodiment of the first aspect by executing the computer instructions.

[0017] According to a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the method described in the first aspect or any optional embodiment of the first aspect.

[0018] The technical solution provided by this application dynamically receives feedback messages from multiple access nodes that are connected to the target QoS directory through a metadata service node, thereby determining the access node currently in the read-write state through the read-write information in the feedback message, and mapping the read-write information of the access node to the feedback information mapping table, dynamically updating the mapping entries corresponding to each access node in the mapping table, so that the metadata service node adjusts the traffic allocation value of each access node in each cycle according to the read-write status of the access node recorded in the table, so that access nodes with more traffic use obtain more allocation values, and access nodes with less traffic use obtain less allocation values. Including multiple metadata service nodes in the distributed file system can share traffic pressure and avoid performance bottlenecks of central nodes. At the same time, dynamically allocating traffic to each access node can improve the efficiency of traffic allocation.

[0019] Furthermore, by recycling excess traffic from unsaturated access nodes and distributing it according to the traffic usage ratio among saturated access nodes, the allocation results are more reasonable. Furthermore, by combining traffic forecasts from multiple previous cycles to predict the access node's traffic value for the next cycle and then calculating the traffic usage ratio, the allocation results are further refined, significantly improving the rationality of traffic distribution.

[0020] It can be seen that the technical solutions provided by the above embodiments of the present application can reduce the system load pressure, improve the balance of flow distribution, and thus improve the efficiency of flow distribution. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The features and advantages of the present invention will be more clearly understood by referring to the accompanying drawings, which are schematic and should not be construed as limiting the present invention in any way. In the accompanying drawings:

[0022] Figure 1 A schematic diagram showing the steps of a QoS directory flow control method in a distributed file system according to one embodiment of the present invention is shown;

[0023] Figure 2 A schematic diagram showing a flow chart of a QoS directory flow control method in a distributed file system according to one embodiment of the present invention is shown;

[0024] Figure 3 A schematic structural diagram of a QoS directory flow control device in a distributed file system according to one embodiment of the present invention is shown;

[0025] Figure 4 A schematic structural diagram of a QoS directory flow control device in a distributed file system according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0026] To make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of the present invention.

[0027] The traffic purchased by the user is the bandwidth and IOPS configuration value, that is, the bandwidth and IOPS upper limit. In a distributed file, after the user purchases the traffic, the traffic arrives from multiple connections, and these connections are on different access nodes. The traffic of multiple access nodes may not be evenly distributed. If the QoS allocation value is not adjusted, the allocation value of some access nodes will be used less, and the allocation value of some access nodes will be saturated. Ultimately, the total configuration value purchased cannot be fully utilized. In the current industry distributed file system, the directory-level QoS implementation depends on the directory-associated QoS configuration and token bucket structure. As mentioned in the background technology, the key points of implementing QoS restrictions are: (1) the access node requests a token from the central node, and the central node issues the token; (2) the access node can cache the token to implement IO traffic restrictions. The above typical solutions have multiple problems: (1) dependence on the central node, there is a performance bottleneck; (2) if the token cache is too small, it will have a greater impact on IO performance; (3) if the token cache is too large, the QoS configuration is updated slowly, and the QoS limit may be exceeded.

[0028] This solution implements a directory-level QoS traffic balancing method in a distributed file system. Multiple access nodes report the traffic status of the QoS directory to the corresponding metadata service node according to the metadata ownership of the QoS directory. QoS values ​​are dynamically allocated according to the traffic status, and QoS allocation values ​​can be adjusted in a timely manner. This achieves dynamic balancing of QoS allocation, improves traffic allocation efficiency, and avoids central node performance bottlenecks and other problems (such as slow QoS changes taking effect, inability to dynamically balance, QoS overlimits, and reliance on manual adjustments by administrators).

[0029] See also Figure 1 and Figure 2 In one embodiment, a QoS directory flow control method in a distributed file system specifically includes the following steps:

[0030] Step S101: Receive the feedback message reported by the access node, and the feedback message includes the read and write information of the access node reading and writing the target QoS directory. Specifically, the embodiment of the present invention receives the feedback message of the access node through multiple metadata service nodes in the distributed file storage system, and then dynamically allocates traffic to each access node according to the read and write information in the feedback message. The metadata service node is responsible for managing the metadata of the distributed file system and providing services for accessing the metadata. There can be multiple active metadata services in the distributed file system, and the metadata service nodes are evenly distributed in the distributed file system (metadata is also called intermediary data or relay data, which is data that describes data, mainly information that describes data attributes, used to support functions such as indicating storage location, historical data, resource search, file records, etc.). The traffic distribution work of different directories is managed separately by the metadata service nodes, which balances the system pressure and avoids the performance bottleneck of the central node.

[0031] Step S102: Based on the feedback message, the metadata service node updates the mapping entry for the access node in the feedback message mapping table of the QoS directory. The mapping entry represents the mapping relationship between the node identifier of the access node and the read / write information. Specifically, after receiving the feedback message from each access node, the metadata service node updates the feedback read / write information into the feedback message mapping table. The feedback information mapping table includes not only the current read / write information of each access node, but also historical read / write information from previous cycles, facilitating calculation of the specific read / write information for each cycle. In this embodiment of the present invention, the read / write information includes the read / write status of each access node and the accumulated traffic value from the initial cycle to the current cycle. The read / write status and accumulated traffic value are used to determine which access nodes are actively reading / writing and whether the traffic of the access nodes currently reading / writing is sufficient. Dynamic traffic redistribution is then performed in each cycle to ensure balanced traffic across all access nodes. Compared to existing technologies, this avoids the uneven traffic distribution problem caused by token bucket-based traffic restriction. Furthermore, when the QoS configuration value changes, the metadata service node can also promptly adjust the current QoS allocation value for each access node in each cycle, solving the problem of slow QoS configuration value changes taking effect compared to traditional token bucket technology. Storing and summarizing feedback information in the form of a mapping table improves the system's efficiency in unified feedback information management and facilitates statistics and calculations. In this embodiment of the present invention, the content of the feedback message forms a mapping entry, which is added to the feedback message table feedbacks_map in a map structure or updates an existing entry in feedbacks_map. The feedback entry includes a key and a value. The key is the access node ID, and the value includes multiple fields: stamp, which is the timestamp when the feedback message is sent; io_stat, which is the read and write status value of the QoS directory; bstats, which is the cumulative statistics of data blocks read and written under the QoS directory; and iostats, which is the cumulative statistics of I / O counts read and written under the QoS directory.

[0032] Step S103: Based on each mapping entry in the feedback message mapping table, QoS configuration information is generated for each access node corresponding to the QoS directory, and the generated QoS configuration information is sent to the corresponding access node to allocate the traffic of the target QoS directory to each access node. Specifically, the metadata service node determines the read and write status of each access node through the feedback information table. For access nodes without read and write, there is no need to waste resources to allocate traffic. For access nodes with read and write, the node analyzes whether its previous QoS allocation value is sufficient, and redistributes traffic to insufficient access nodes in the next cycle. The generated QoS configuration information is promptly sent to each access node to ensure that each access node uses the most appropriate traffic in each cycle, thereby improving the overall efficiency of the traffic purchased by users.

[0033] Specifically, in one embodiment, the above step S102 specifically includes the following steps:

[0034] Step S1021: If the feedback message reported by the access node indicates that the access node has not performed any read or write behavior within a preset number of periods, the mapping entry corresponding to the access node is removed from the feedback message mapping table.

[0035] Step S1022: If the feedback message reported by the access node indicates that the access node has performed a read or write behavior for the first time, a new mapping entry is added to the feedback message mapping table based on the feedback message.

[0036] Step S1023: At the end of each cycle, the traffic accumulation value of the access node corresponding to each mapping entry is counted, and each mapping entry in the feedback message mapping table is modified based on the counted traffic accumulation value.

[0037] Specifically, in an embodiment of the present invention, a status bit is set in the QoS structure (a QoS data structure independent of the QoS directory, with a one-to-one relationship with the QoS directory's metadata cache) to indicate the read and write status of files in the QoS directory (a directory configured with QoS attributes). Files in the QoS directory have three read and write states: NOIO, FIRST_IO, and IO. The NOIO state indicates that no files in the QoS directory are being read or written. Only access operations to the directory or subdirectories, as well as access operations to metadata of files within the directory, such as readdir and stat, are occurring. To ensure convenient system statistics on read and write status, mapping entries for access nodes without read or write access nodes are removed from the feedbacks_map. For any access node, if no read or write occurs within a preset K (K is a positive integer) cycles, the access node performs a state transition from IO to NOIO. Specifically, the feedbacks_map is traversed in each cycle to check whether a feedback entry exists whose timestamp is K cycles behind the current time. If such a mapping entry exists, the entry is removed from the feedbacks_map. The FIRST_IO state indicates that the reading and writing of files in the QoS directory has occurred for the first time, or that the reading and writing of files has not occurred for several cycles before, and the reading and writing of files in the QoS directory have been restored. The FIRST_IO state can only be entered from the NOIO state. When the reading and writing of a certain access node is counted for the first time, the mapping entry of the node will be added to the feedback message mapping table. The IO state indicates that the files in the QoS directory are continuously read and written. The IO state can only be entered from the FIRST_IO state, and the NOIO state can only be entered after the IO state. The mapping entries in the feedback message mapping table correspond to access nodes that are all access nodes with reading and writing, and the subsequent traffic redistribution steps are also performed on these access nodes.

[0038] The timing of the three state transitions is as follows:

[0039] NOIO state transition to FIRST_IO state: After the QoS structure is created and initialized, it is in NOIO state, indicating that no files in the QoS directory are being read or written. When any file is counted as being read or written, the state of the QoS structure is changed from NOIO to FIRST_IO state, and at the same time, the access node sends a FIRST_IO type feedback message to the metadata service node where the QoS directory metadata is located; FIRST_IO state transition to IO state: When files in the QoS directory continue to be read and written, the state of the associated QoS structure changes from FIRST_IO to IO state; IO state transition to NOIO state: When a preset number of cycles have passed and no files are detected being read or written in the QoS directory, the QoS structure changes from IO state to NOIO state, and at the same time, the access node sends a NOIO type feedback message to the metadata service node where the QoS directory metadata is located. When the state remains in the IO state, the access node will also periodically send IO type feedback messages to report the cumulative number of read and write times and the cumulative data block size at the end of the current cycle, and store the cumulative number of read and write times and the cumulative data block size counted each time in the corresponding mapping entry, so that each mapping entry contains the cumulative number of read and write times and the cumulative data block size of all historical cycles. At the access node of the distributed file storage cluster, the method for counting the read and write traffic at the QoS directory level is: if the read and write files are associated with the QoS structure, the cumulative number of read and write times and the cumulative data block size of the read and write IO are saved on the QoS structure. Within the read and write interface, the number of read and write times will be counted and saved in the cumulative number of read and write statistics of the QoS structure. Each time the read (or write) interface is called, the corresponding statistical count will be increased by one; in addition, the read and write block size will be counted and saved in the read and write block size statistics of the QoS structure. Each time a read (or write) is successful, the size of the read (or written) data block will be accumulated in the data block statistics value. When the cumulative count of read and write times or the cumulative value of data blocks increases, it is determined that there is read and write in this cycle; otherwise, it is determined that there is no read and write in this cycle. Through the above processing, it is ensured that the mapping entries in the feedback message mapping table are all information about read and write access nodes, which improves the efficiency of subsequent traffic calculation and statistics and avoids errors.

[0040] Specifically, in one embodiment, the above step S103 specifically includes the following steps:

[0041] Step S1031: If the current cycle is the initial cycle or the cycle for new node joining, the QoS configuration value is evenly distributed to the access nodes corresponding to each mapping entry, and the distribution result is used as the QoS configuration information corresponding to each access node.

[0042] Specifically, the steps for generating QoS configuration information provided by the embodiment of the present invention are specifically divided into two parts. In order to improve the balance of traffic distribution of each access node, the unused traffic portion of the access node whose actual traffic usage does not reach the QoS allocation value, that is, the excess allocation amount, is recovered, and then the recovered traffic is redistributed to the access node whose actual traffic usage has reached the QoS allocation value. However, to determine whether the actual traffic usage of the access node has reached the QoS allocation value, the historical read and write information of the access node needs to be analyzed. When the current cycle is the initial cycle or the cycle in which a new node joins, there are nodes without historical read and write information, and then traffic cannot be recovered based on the historical read and write information. At this time, the system automatically distributes the traffic evenly and evenly distributes the QoS configuration value of the directory to all access nodes that are currently in the read and write state. The specific distribution formula is as follows:

[0043]

[0044] Among them, iops_alloc is the IOPS allocation value for any access node; bw_alloc is the bandwidth allocation value for any access node; iops_limit is the IOPS limit value configured by the user for the QoS directory; bw_limit is the bandwidth limit value configured by the user for the QoS directory; num_clients: the number of access nodes performing reading and writing.

[0045] Step S1032: If the current cycle is not the initial cycle or the cycle for new node joining, the saturation state of the access node corresponding to each mapping entry is determined.

[0046] Step S1033: Reclaiming the excess allocation amount of each unsaturated access node and reallocating the excess allocation amount among each saturated access node to generate QoS configuration information of each access node in the next period.

[0047] Specifically, due to external or internal reasons, the traffic usage of access nodes is not fixed. An access node with high traffic usage in the current cycle may experience lower traffic usage in the next cycle. The QoS allocation value (allocated traffic cap) assigned to each access node may result in traffic usage far below the QoS allocation value due to low access node usage. Therefore, dynamically adjusting the traffic allocation for these undersaturated access nodes can maximize the efficiency of QoS configuration value utilization. Therefore, it is necessary to calculate the saturation status of each access node for subsequent processing. In this embodiment of the present invention, actual traffic statistics are achieved by using a metadata service node to count the number of read and write operations and the cumulative data block size of each access node during each cycle. The actual traffic value is then compared with the QoS allocation value of each access node to determine the corresponding saturation status. Unused traffic from each undersaturated access node is then recycled and reallocated to other saturated access nodes. This ensures that the resources purchased by users are fully utilized. Traffic is dynamically allocated within each cycle based on access node usage needs, avoiding issues such as slow QoS change implementation, inability to dynamically balance, QoS overruns, and reliance on manual administrator adjustments in existing technologies.

[0048] Specifically, in one embodiment, the above step S1032 specifically includes the following steps:

[0049] Step S10321: Calculate the traffic value of each access node in the previous cycle based on the accumulated traffic value to obtain multiple first traffic values. Specifically, in this embodiment, whether each access node is a saturated access node is determined by the relationship between the allocation value and the actual traffic usage value of each access node in the previous cycle. The calculation formula for the actual traffic usage value (i.e., the first traffic value) of the access node in the previous cycle is as follows:

[0050] iops_stats i =(new_op_sum i -last_op_sum i ) / T,i∈[0,N)

[0051] Among them, iops_stats i Indicates the traffic value of access node i in the last cycle (i.e., the first traffic value), new_op_sum i Indicates the cumulative traffic value of access node i at the end of the previous cycle, last_op_sum i It represents the cumulative traffic value of access node i at the end of the previous cycle, T represents the time of a cycle (unit: second), and N represents the number of access nodes.

[0052] Step S10322: Compare the first traffic value with the QoS allocation value of the previous period. If the first traffic value is less than the QoS allocation value of the previous period, determine that the access node is an unsaturated access node.

[0053] Step S10323: If the first traffic value is equal to the QoS allocation value of the previous period, it is determined that the access node is a saturated access node.

[0054] Specifically, when the first traffic value of any access node in the previous cycle is less than the QoS allocation value of the previous cycle, the node is an unsaturated access node; when the first traffic value of any access node in the previous cycle is equal to the QoS allocation value of the previous cycle, the node is a saturated access node. In actual situations, there are a certain number of access nodes whose actual traffic usage values ​​are very close to their QoS allocation values. Therefore, the relationship between the actual traffic usage value and the QoS allocation value in this embodiment is not completely equal, but approximately equal. When the first traffic value reaches a preset proportion of the QoS allocation value of the previous cycle, they are considered equal. For example, when the first traffic value reaches 95% of the QoS allocation value, it is considered equal, and the access node is a saturated access node. This avoids ignoring the situation where these access nodes are saturated access nodes, and improves the accuracy of node saturation status recognition.

[0055] Specifically, in one embodiment, based on steps S10321 to S10323, step S1033 specifically includes the following steps:

[0056] Step S10331: Calculate the allocation increment, where the allocation increment is the ratio of the sum of the excess allocations of all unsaturated access nodes to the number of saturated access nodes.

[0057] Step S10332: Add the QoS allocation value of the saturated access node in the previous period to the allocation increment to generate a new QoS allocation value of each saturated access node in the next period.

[0058] Step S10333: The traffic value of each unsaturated access node in the previous period is used as the new QoS allocation value of each unsaturated access node in the next period.

[0059] Specifically, in step S1032, the difference between the QoS allocation value and the first traffic value of each unsaturated access node in the previous cycle is calculated. The difference represents the unused traffic allocation (excess allocation) of the unsaturated access node in the previous cycle. The metadata node then reclaims this portion of the traffic allocation value. The sum of all the differences is the total traffic reclaim amount. Then, in this embodiment, the total reclaimed amount is evenly distributed to each saturated access node. Specifically, the total reclaimed amount is divided by the number of saturated access nodes to obtain an allocation increment value. This allocation increment value is then distributed to each saturated access node. The sum of the allocation increment and the QoS allocation value of each saturated access node in the previous cycle is the new QoS allocation value of the saturated access node in the next cycle. The first traffic value actually used by the unsaturated access node in the previous cycle is the new QoS allocation value of the unsaturated access node in the next cycle. Through these steps, the excess allocation of the unsaturated access node is redistributed to the saturated access node in each cycle, ensuring traffic balance among the access nodes and improving the utilization rate of the purchased traffic by users.

[0060] Specifically, in another embodiment, the above steps S1032 to S1033 specifically include the following steps:

[0061] Step S10335: Based on the change trend of the traffic accumulation value of each access node in multiple cycles, the QoS allocation value of each access node in the next cycle is predicted to obtain multiple traffic prediction values.

[0062] Step S10336: Determine whether the access nodes are saturated or unsaturated based on the relationship between the traffic prediction value and the QoS allocation value of the previous period.

[0063] Specifically, in this embodiment, in order to further improve the accuracy of traffic redistribution, not only the first traffic value and QoS allocation value of each access node in the previous cycle are used to determine the saturation state of each access node, but the historical traffic accumulation value information of each access node is used to calculate the traffic value of each cycle of the access node in the previous M (M is a positive integer) cycles, and the traffic value of each historical cycle is input into the linear regression function as an array as a whole, so as to predict the traffic usage of the access node in the next cycle and obtain the traffic prediction value. This further improves the accuracy of the estimation of the traffic usage of the access node in the next cycle. The linear regression function prediction is an existing technology and will not be repeated in this embodiment. Afterwards, the saturation state of the access node is determined by judging the relationship between the traffic prediction value of each access node and the QoS allocation value of the previous cycle.

[0064] Step S10337: Calculate the ratio of the traffic prediction value of each saturated access node to the total traffic prediction value of the saturated access nodes to obtain the split ratio.

[0065] Step S10338: Split all excess allocations into multiple allocation increments based on the split ratio and allocate them to corresponding saturated access nodes. The QoS allocation value of the saturated access node in the next cycle is the sum of the QoS allocation value of the previous cycle and the allocation increment.

[0066] Step S10339: The predicted traffic value of the unsaturated access node is used as the QoS allocation value of the unsaturated access node in the next period.

[0067] Specifically, unsaturated access nodes do not participate in the redistribution of the total recycled capacity; only saturated access nodes participate. The ratio of each saturated access node's predicted traffic volume to the total predicted traffic volume of saturated access nodes is calculated, and the total recycled capacity is divided according to the obtained ratio. For example, if the IPOS traffic predictions of three saturated access nodes are 102 times / second, 120 times / second, and 150 times / second, respectively, and the total recycled capacity is 1000 times / second, the allocation increment for the first saturated access node is: 1000 × 102 / (102 + 120 + 150) times / second. The calculation method for the other two saturated access nodes is similar. This further improves the accuracy of the redistribution of the total recycled capacity among saturated access nodes, ensuring more balanced QoS directory traffic distribution. The QoS allocation value for a saturated access node in the next cycle is the sum of the predicted traffic volume and the allocation increment calculated in the above steps. The QoS allocation value for an unsaturated access node in the next cycle is its predicted traffic volume. This ensures that the QoS directory configuration values ​​are used to the maximum extent by each access node, improving traffic balance.

[0068] Specifically, in one embodiment, step S10336 includes the following steps:

[0069] Step S61: Compare the traffic prediction value of the access node with the QoS allocation value of the previous period. If the traffic prediction value is less than the QoS allocation value of the previous period, determine that the access node is an unsaturated access node.

[0070] Step S62: If the traffic prediction value is equal to the QoS allocation value of the previous period, the access node is determined to be a first saturated access node.

[0071] Step S63: If the traffic prediction value is greater than the QoS allocation value of the previous period, the access node is determined to be a second saturated access node.

[0072] Step S64: for the first saturated access node, increase its traffic prediction value to a second traffic prediction value according to a preset ratio.

[0073] Specifically, when the traffic prediction value of an access node is less than the QoS allocation value of its previous cycle, it indicates that the node is predicted to be unable to fully utilize the previous QoS allocation value, and thus the access node is an unsaturated access node. When the traffic prediction value of an access node is equal to the QoS allocation value of its previous cycle, it indicates that the node is predicted to fully utilize the previous QoS allocation value, and the node is marked as a first saturated access node. When the traffic prediction value of an access node is greater than the QoS allocation value of its previous cycle, it indicates that the node is predicted to not only fully utilize the previous QoS allocation value, but also that the previous QoS allocation value is not enough for the access node to use, and thus the node is marked as a second saturated access node. The saturation status of each access node is further clarified. In order to prevent the redistribution ratio of the first saturated access node to all saturated access nodes from being too low, resulting in the allocation increment of the first saturated access node being too small, and the traffic exceeding the limit in the next cycle. First, the traffic allocation value of the first saturated access node in the next cycle is preprocessed, and the traffic prediction value of the first saturated access node is increased to a second traffic prediction value according to a preset ratio. For example, the IPOS prediction value of the first saturated access node is 50 times / second, which is increased by 4% to a second traffic prediction value of 52 times / second.

[0074] Through the above steps, the excess allocation of unsaturated access nodes determined based on traffic prediction values ​​is redistributed to saturated access nodes in each cycle, ensuring traffic balance among access nodes and improving the utilization rate of user-purchased traffic.

[0075] Specifically, in one embodiment, after step S1033, the method for controlling QoS directory traffic in a distributed file system provided by an embodiment of the present invention further includes the following steps:

[0076] Step S104: When the QoS configuration value is adjusted downward, the new QoS allocation value of each access node in the next period is adjusted downward simultaneously according to the adjustment ratio of the QoS configuration value.

[0077] Step S105: When the QoS configuration value is increased, the ratio of the new QoS allocation value of each saturated access node in the next period to the total QoS allocation value of the saturated access node is calculated to generate a second split ratio.

[0078] Step S106: allocating the increase amount of the QoS configuration value to the corresponding saturated access nodes based on the second split ratio.

[0079] Specifically, when the configuration value of the QoS catalog is overall lowered, the new allocation value will be synchronously lowered according to the ratio of the configuration value reduction based on the new QoS allocation value of each access node in the next cycle. Compared with the existing technology, the real-time adjustment of the configuration value is guaranteed, and the slow response of each access node to the configuration value change is avoided. For example, there are currently two access nodes. If the QoS configuration value is lowered by 2%, the QoS allocation value of each access node will be lowered by 2%. When the configuration value of the QoS catalog is raised, it is slightly different from the reduction. The allocation value of unsaturated access nodes does not need to be changed because of their weak traffic usage capabilities. It is sufficient to increase the allocation value only for saturated access nodes, but the specific increase amount needs to be determined according to the ratio of the allocation value of each saturated access node to the total allocation value (i.e., the second split ratio). For example, if the IPOS value in the QoS configuration is increased by 2%, the specific increase of 2% is calculated to be 1000 times / second. There are currently two saturated access nodes. The latest allocation value of saturated access node A is A, and the latest allocation value of saturated access node B is B. Then the specific increase in the configuration value of saturated access node A is 1000×A / (A+B). The same applies to the increase in the configuration value of saturated access node B. Note that in steps S61 to S64, the QoS configuration value increase needs to be performed on both the first saturated node and the second saturated access node. By increasing the configuration value of the saturated access node, this embodiment not only increases the speed of configuration value changes, but also further improves the balance of traffic distribution.

[0080] Specifically, in one embodiment, after step S103, the method for controlling QoS directory traffic in a distributed file system provided by the embodiment of the present invention further includes the following steps:

[0081] Step S107: Based on the QoS configuration information, a second mapping entry of the access node is updated in the traffic distribution mapping table of the QoS directory, where the second mapping entry is used to represent the mapping relationship between the node identifier of the access node and the QoS configuration information.

[0082] Step S108: converting the changed second mapping entry in the traffic distribution mapping table into a traffic distribution message and sending the message to the corresponding access node.

[0083] Specifically, this embodiment stores QoS configuration information in a traffic allocation mapping table, ensuring that all entries in the table represent access nodes participating in traffic allocation, facilitating management and statistics. Traffic allocation value change messages are only sent to access nodes corresponding to the changed entries in the table, reducing the number of new messages sent by the system and improving efficiency. In this embodiment, the traffic allocation mapping table, allocations_map, utilizes a map structure and can be accessed through the QoS directory metadata structure inode index. Each QoS directory metadata corresponds to an allocations_map. The allocations_map contains several allocation entries, each corresponding to the allocation information of an access node. The allocation entry contains a key and a value. The key is the access node ID, i.e., the ID of the NAS head that assigns the allocation value (the NAS head is the external access node for distributed file storage and serves as the feedback processing entry for distributed file system traffic). The value includes multiple fields: iops_alloc, i.e., the IOPS value allocated to the corresponding access node; bw_alloc, i.e., the bandwidth value allocated to the corresponding access node. Based on the changed entries in the allocations_map, the metadata service node sends the corresponding allocations_notify notification message to the corresponding access node.

[0084] Through the above steps, the technical solution provided by the present application dynamically receives feedback messages from multiple access nodes connected to the target QoS directory through the metadata service node, thereby judging the access node currently in the read-write state through the read-write information in the feedback message, and mapping the read-write information of the access node to the feedback information mapping table, and dynamically updating the mapping entries corresponding to each access node in the mapping table, so that the metadata service node adjusts the traffic allocation value of each access node in each cycle according to the read-write status of the access node recorded in the table, so that the access node with more traffic usage obtains more allocation value, and the access node with less traffic usage obtains less allocation value. Including multiple metadata service nodes in the distributed file system can share the traffic pressure and avoid the performance bottleneck of the central node. Dynamically allocating traffic to each access node solves the problem of uneven traffic distribution in the existing technology.

[0085] Furthermore, by recycling excess traffic from unsaturated access nodes and distributing it according to the traffic usage ratio among saturated access nodes, the allocation results are more reasonable. Furthermore, by combining traffic forecasts from multiple previous cycles to predict the access node's traffic value for the next cycle and then calculating the traffic usage ratio, the allocation results are further refined, significantly improving the rationality of traffic allocation.

[0086] It can be seen that the technical solutions provided by the above embodiments of the present application can reduce the system load pressure, improve the balance of traffic distribution, and solve the problems of QoS central node bottleneck and unbalanced QoS directory traffic distribution.

[0087] like Figure 3 As shown, this embodiment also provides a QoS directory flow control device in a distributed file system, the device comprising:

[0088] The information receiving module 101 is configured to receive feedback messages reported by access nodes, including read and write information of the access nodes on the target QoS directory. Detailed information can be found in the description of step S101 in the above method embodiment, which will not be repeated here.

[0089] Feedback message mapping module 102 is configured to update a mapping entry for the access node in a feedback message mapping table of the target QoS directory based on the feedback message. The mapping entry represents a mapping relationship between the node identifier of the access node and the read / write information. For details, see the description of step S102 in the above method embodiment and will not be repeated here.

[0090] Traffic configuration module 103 is configured to generate QoS configuration information for each access node corresponding to the target QoS category based on each mapping entry in the feedback message mapping table, and to distribute the generated QoS configuration information to the corresponding access node to allocate traffic in the target QoS category to each access node. For details, see the description of step S103 in the above method embodiment and will not be repeated here.

[0091] The QoS directory flow control device in the distributed file system provided by the embodiment of the present invention is used to execute the QoS directory flow control method in the distributed file system provided by the above embodiment. Its implementation method and principle are the same. For details, please refer to the relevant description of the above method embodiment and will not be repeated here.

[0092] Through the collaborative cooperation of the above-mentioned components, the QoS directory traffic control device in the distributed file system provided by the present application dynamically receives feedback messages from multiple access nodes connected to the target QoS directory through the metadata service node, thereby judging the access node currently in the read-write state through the read-write information in the feedback message, and mapping the read-write information of the access node to the feedback information mapping table, and dynamically updating the mapping entries corresponding to each access node in the mapping table, so that the metadata service node adjusts the traffic allocation value of each access node in each cycle according to the read-write status of the access node recorded in the table, so that the access node with more traffic usage obtains more allocation value, and the access node with less traffic usage obtains less allocation value. Including multiple metadata service nodes in the distributed file system can share the traffic pressure and avoid the performance bottleneck of the central node. Dynamically allocating traffic to each access node solves the problem of uneven traffic distribution in the prior art.

[0093] Furthermore, by recycling excess traffic from unsaturated access nodes and distributing it according to the traffic usage ratio among saturated access nodes, the allocation results are more reasonable. Furthermore, by combining traffic forecasts from multiple previous cycles to predict the access node's traffic value for the next cycle and then calculating the traffic usage ratio, the allocation results are further refined, significantly improving the rationality of traffic allocation.

[0094] It can be seen that the technical solutions provided by the above embodiments of the present application can reduce the system load pressure, improve the balance of traffic distribution, and solve the problems of QoS central node bottleneck and unbalanced QoS directory traffic distribution.

[0095] Figure 4 The present invention shows a QoS directory flow control device in a distributed file system according to an embodiment of the present invention. The device includes a processor 901 and a memory 902, which can be connected via a bus or other means. Figure 4 The bus connection is taken as an example.

[0096] The processor 901 may be a central processing unit (CPU). The processor 901 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above chips.

[0097] Memory 902, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer executable programs, and modules, such as the program instructions / modules corresponding to the methods in the above-described method embodiments. Processor 901 executes the non-transitory software programs, instructions, and modules stored in memory 902 to perform various processor functions and data processing, thereby implementing the methods in the above-described method embodiments.

[0098] The memory 902 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created by the processor 901, etc. In addition, the memory 902 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 902 may optionally include a memory remotely located relative to the processor 901, and these remote memories may be connected to the processor 901 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0099] One or more modules are stored in the memory 902 and, when executed by the processor 901 , perform the method in the above method embodiment.

[0100] The specific details of the QoS directory traffic control device in the above-mentioned distributed file system can be understood by referring to the corresponding descriptions and effects in the above-mentioned method embodiments, and will not be repeated here.

[0101] Those skilled in the art will appreciate that all or part of the processes in the above-described embodiments can be implemented by instructing related hardware through a computer program. The implemented program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above-described embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk drive (HDD), or a solid-state drive (SSD). The storage medium can also include a combination of the above-mentioned types of memory.

[0102] Although the embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention. Such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A QoS directory flow control method in a distributed file system, characterized in that: The method is applied to a metadata service node, and includes: receiving a feedback message reported by an access node, wherein the feedback message includes read and write information of the access node on a target QoS directory; Based on the feedback message, updating a mapping entry of the access node in a feedback message mapping table of the target QoS directory, where the mapping entry is used to represent a mapping relationship between a node identifier of the access node and the read / write information; Generate QoS configuration information for each access node corresponding to the target QoS directory according to each mapping entry in the feedback message mapping table, and send the generated QoS configuration information to the corresponding access node to distribute the traffic of the target QoS directory to each access node; The updating of the mapping entry of the access node in the feedback message mapping table of the target QoS directory based on the feedback message includes: if the feedback message reported by the access node indicates that the access node has not performed any read or write behavior within a preset number of cycles, removing the mapping entry corresponding to the access node from the feedback message mapping table; if the feedback message reported by the access node indicates that the access node has performed read or write behavior for the first time, adding a new mapping entry to the feedback message mapping table based on the feedback message; wherein, at the end of each cycle, the cumulative traffic value of the access node corresponding to each mapping entry is counted, and each mapping entry in the feedback message mapping table is modified based on the counted cumulative traffic value.

2. The method according to claim 1, characterized in that Generating QoS configuration information for each access node corresponding to the target QoS directory according to each mapping entry in the feedback message mapping table includes: If the current cycle is the initial cycle or the cycle in which a new node joins, the QoS configuration value is evenly distributed to the access nodes corresponding to each mapping entry, and the distribution result is used as the QoS configuration information corresponding to each of the access nodes; If the current cycle is not an initial cycle or a cycle in which a new node joins, determining a saturation state of an access node corresponding to each mapping entry; The excess allocation amount of each unsaturated access node is recovered, and the excess allocation amount is redistributed among each saturated access node to generate QoS configuration information of each access node in the next period.

3. The method according to claim 2, characterized in that The determining the saturation state of the access node corresponding to each mapping entry includes: Calculate the traffic value of each access node in the previous period based on the traffic accumulation value to obtain multiple first traffic values; comparing the first traffic value with a QoS allocation value in a previous period, and if the first traffic value is less than the QoS allocation value in the previous period, determining that the access node is an unsaturated access node; If the first traffic value is equal to the QoS allocation value of the previous period, it is determined that the access node is a saturated access node.

4. The method according to claim 3, characterized in that The recovering of the excess allocation amount of each unsaturated access node and redistributing the excess allocation amount among each saturated access node includes: Calculating an allocation increment, where the allocation increment is a ratio of a sum of excess allocations of all unsaturated access nodes to the number of saturated access nodes; Adding the QoS allocation value of the saturated access node in the previous period to the allocation increment to generate a QoS allocation value of each saturated access node in the next period; The traffic value of each unsaturated access node in the previous period is used as the QoS allocation value of each unsaturated access node in the next period.

5. The method according to claim 2, characterized in that The determining of the saturation status of the access nodes corresponding to the respective mapping entries, recovering the excess allocation amount of the respective unsaturated access nodes and redistributing the excess allocation amount among the respective saturated access nodes comprises: Based on the changing trend of the accumulated traffic value of each access node in multiple cycles, the QoS allocation value of each access node in the next cycle is predicted to obtain multiple traffic prediction values; Determining saturated access nodes and unsaturated access nodes among the access nodes according to a magnitude relationship between the traffic prediction value and the QoS allocation value of the previous period; Calculate the ratio of the traffic prediction value of each saturated access node to the total traffic prediction value of the saturated access nodes to obtain the split ratio; Splitting all excess allocations into multiple allocation increments based on the split ratio, and allocating the increments to corresponding saturated access nodes, wherein the QoS allocation value of the saturated access node in the next cycle is the sum of the QoS allocation value of the previous cycle and the allocation increment; The predicted traffic value of the unsaturated access node is used as the QoS allocation value of the unsaturated access node in the next period.

6. The method according to claim 5, characterized in that The determining, according to a magnitude relationship between the traffic prediction value and the QoS allocation value of the previous period, a saturated access node and an unsaturated access node among the access nodes includes: comparing a traffic prediction value of the access node with a QoS allocation value in a previous period, and determining that the access node is an unsaturated access node if the traffic prediction value is less than the QoS allocation value in the previous period; If the traffic prediction value is equal to the QoS allocation value of the previous period, determining that the access node is a first saturated access node; If the traffic prediction value is greater than the QoS allocation value of the previous period, determining that the access node is a second saturated access node; For the first saturated access node, its traffic prediction value is increased to a second traffic prediction value according to a preset ratio.

7. The method according to claim 4 or 5, characterized in that The method further comprises: When the QoS configuration value is lowered, the QoS allocation value of each access node for the next cycle is lowered simultaneously according to the reduction ratio of the QoS configuration value; When the QoS configuration value is increased, the ratio of the QoS allocation value of each saturated access node in the next cycle to the total QoS allocation value of the saturated access node is calculated to generate a second split ratio; The increase amount of the QoS configuration value is distributed to the corresponding saturated access nodes based on the second split ratio.

8. The method according to claim 1, characterized in that The sending the generated QoS configuration information to the corresponding access node includes: Based on the QoS configuration information, updating a second mapping entry of the access node in the traffic distribution mapping table of the target QoS directory, where the second mapping entry is used to represent a mapping relationship between the node identifier of the access node and the QoS configuration information; Generate a traffic allocation message corresponding to the changed mapping entry in the traffic allocation mapping table, and send the traffic allocation message to the corresponding access node.

9. A QoS directory flow control device in a distributed file system, characterized in that: The device is applied to a metadata service node, and includes: An information receiving module, configured to receive a feedback message reported by an access node, wherein the feedback message includes read and write information of the access node reading and writing the target QoS directory; A feedback message mapping module is configured to update, in a feedback message mapping table of the target QoS directory, a mapping entry of the access node based on the feedback message, the mapping entry being used to characterize a mapping relationship between a node identifier of the access node and the read / write information; updating, in the feedback message mapping table of the target QoS directory, the mapping entry of the access node based on the feedback message comprises: if the feedback message reported by the access node characterizes that the access node has not undergone any read / write behavior within a preset number of cycles, removing the mapping entry corresponding to the access node from the feedback message mapping table; if the feedback message reported by the access node characterizes that the access node has undergone any read / write behavior for the first time, adding a new mapping entry to the feedback message mapping table based on the feedback message; wherein, at the end of each cycle, the accumulated traffic value of the access node corresponding to each mapping entry is counted, and each mapping entry in the feedback message mapping table is modified based on the counted accumulated traffic value; The traffic configuration module is used to generate QoS configuration information for each access node corresponding to the target QoS directory according to each mapping entry in the feedback message mapping table, and send the generated QoS configuration information to the corresponding access node to distribute the traffic of the target QoS directory to each access node.

Images