CN105610921A - Erasure code filing method based on data cache in cluster - Google Patents

Abstract

The invention discloses an erasure code archiving method based on data cache under a cluster, which comprises the following steps: (1) according to a user access request, read the required data blocks from the production node where it is located to the node buffer, and The data block is sent to the access node, and forwarded to the archiving node according to the archive data status table; (2) the archiving node updates the number of data blocks of each data node according to the received data block; judges whether the archiving node is All the data blocks of the current stripe are received, if so, the current stripe is archived; if not, the cold data block is read from the production cluster, and the stripes are assembled and then archived; the correction provided by the present invention Compared with the common erasure code archiving method, the erasure code archiving method can greatly reduce the number of times the archiving node reads data blocks from the production cluster, without affecting user access. Perform erasure code archiving operations under certain circumstances to improve archiving efficiency.

Description

A cluster-based erasure code archiving method based on data caching

technical field

The invention belongs to the technical field of computer storage, and more specifically relates to an erasure code archiving method based on data caching under a cluster.

Background technique

In today's information age, the amount of data is increasing rapidly, and large-scale storage clusters have been more and more widely used. According to the frequency of data access, the data in the storage cluster can be divided into hot data (HotData), warm data (WarmData) and cold data (ColdData) in turn. As the system runs longer, the access frequency of hot data will decrease, become warm data, and finally become cold data. Hot data is usually stored in the production cluster (ProductionCluster) in three or more copies. A method to improve the utilization rate of the production cluster system is to migrate the copy data with low access frequency to the erasure-coded archive cluster (ArchivalCluster), which uses RS codes to organize the data stored in it; the erasure-coded archive (Erasure- codedArchival) refers to the operation of migrating data from replica storage to erasure code storage.

The existing erasure code archiving and user access to the replica cluster are two independent processes; the existing erasure code archiving includes (1) obtaining the data bitmap of the data to be archived in the production cluster; (2) reading from the production cluster Take the copy data; (3) perform erasure code encoding on the read copy data to generate verification blocks; (4) write a copy of the copy data and corresponding verification blocks into the erasure code archiving cluster; (5) The remaining cold data on the production cluster is recycled, and all data copies in the production cluster that participate in erasure code archiving are deleted. The user's access to the replica cluster includes (1) obtaining the metadata information of the required data block in the production cluster; (2) sending a read request to the node where the data block is located; (3) the node reading the corresponding data block to the node buffer (4) The data node sends the data block and forwards the data block to the client.

For this erasure code archiving, if there are too many requests for data archiving, offline archiving is used, and data access and archiving are independent; however, in the current era of big data, there is almost no dedicated archiving time, and online archiving is required; for online archiving , the nodes in the production cluster always keep hot data and cold data, and the above-mentioned erasure code archiving method has the following problems:

(1) The archiving request is directly sent to the production cluster, and the production cluster is also providing access services for users at this time; the archiving request brings additional external memory access, memory cache occupation, and network resource occupation, resulting in response time for user requests greatly increase;

(2) For archiving tasks, because nodes in the production cluster need to process user access requests and data archiving requests at the same time, there is resource competition between the two; when a storage cluster node has high frequency of user access requests, the node's The archiving request will not be responded for a long time, causing the archiving node to fail to obtain data blocks in time, which will affect the overall archiving efficiency.

Contents of the invention

In view of the above defects or improvement needs of the prior art, the present invention provides an erasure code archiving method based on data caching under a cluster, the purpose of which is to improve the efficiency of erasure code archiving.

In order to achieve the above purpose, according to one aspect of the present invention, a method for archiving erasure codes based on data caching under a cluster is provided, which specifically includes the following steps:

(1) According to the user's access request, read the required data block from the production node where it is located to the node buffer, and send the data block to the access node, and forward it to the archive node according to the archive data status table;

(2) The archiving node updates the number of data blocks of each data node according to the received data blocks; and judges whether the archiving node has received all the data blocks of the current stripe, and if so, archives the current stripe ; If not, read the cold data block from the production cluster, and archive the stripes.

Compared with ordinary erasure code archiving, the above erasure code archiving method can greatly reduce the number of times the archiving node reads data blocks from the production cluster due to the use of data blocks accessed by users in the cache, so that it can be stored in almost no time Perform archiving operations when user access is affected to improve archiving efficiency;

On the other hand, the data blocks of user access cache are forwarded to the archive node, which avoids repeated read requests for the same data block in the production node for user access and common erasure code archiving, which can effectively reduce network transmission in the production cluster The amount of data reduces the load on the entire network.

Preferably, the above-mentioned step (1) specifically includes the following sub-steps:

(1.1) The access node obtains the address of the required data block in the production cluster from the management server;

(1.2) Send a read request to the node corresponding to the address of the above-mentioned data block in the generation cluster;

(1.3) Read the data block from the production node in the generation cluster where it is located to the node buffer;

(1.4) According to the archived data state table, judge whether the data block has been forwarded to the archiving node, if so, then forward the data block to the access node; if not, forward the data block to the access node and the access node at the same time Archiving node; this step directly uses the switch in the cluster to forward the data accessed by the user to the archiving node.

Preferably, the above-mentioned step (2) specifically includes the following sub-steps:

(2.1) The archive node receives data blocks (hot data), and updates the number of data blocks of each data node included in the archive node;

(2.2) The archiving node initializes the current stripe, analyzes the node composition of the stripe, judges which data node the received data block belongs to, and initializes the storage variable; in this step, the node includes a data node and a check node;

(2.3) Judging by the archiving node whether all the data blocks of the current stripe have been received, if so, enter step (2.7); if not, then enter step (2.4);

(2.4) During the time interval of T, the archiving node receives data blocks from the user cache;

(2.5) Judging whether the k data blocks of the current stripe are collected, if so, enter step (2.7); if not, then enter step (2.6); wherein, the k value is equal to the number of data nodes;

(2.6) The archiving node directly reads the cold data from the production cluster to gather μ stripes, and then enters step (2.3);

Among them, the number of data blocks read from data node j is Num _j , Num _j = max{(μ-a _j -λ(j)),0}, μ is the maximum number of blocks that can be archived by the archiving node within time T The number of stripes, a _j is the number of data blocks containing data node j in the archive node, λ(j) represents the number of data blocks that the archive node will receive from data node j in the next time interval T;

(2.7) archive the current strip, and add 1 to the value of the strip number i, and update the archived data status table;

(2.8) judge whether to complete the archiving of all stripes according to the archiving data status table, if not, then enter step (2.3); if so, then end archiving;

In the above process, the archiving node can use the cached data to receive user blocks without affecting user access, which can play a role in improving archiving efficiency.

Generally speaking, compared with the prior art, the above technical solutions conceived by the present invention can achieve the following beneficial effects:

(1) The erasure code archiving method based on data caching under the cluster provided by the present invention uses cached data accessed by users to improve archiving efficiency without affecting user access: by dividing cached data accessed by users into blocks Send it directly to the archiving node. Compared with the ordinary erasure code archiving method, the method provided by the present invention can reduce the number of disk reads, thereby speeding up the production node external storage data block reading, production node memory space occupation, and production node The network card sends the archived data into blocks to improve the efficiency of archiving; only when a stripe is not fully received, it will directly read the production cluster data blocks, and the number of read data blocks and the number of blocks Compared with the amount and frequency of data accessed by users, it is much lower, and basically does not affect user access;

(2) The erasure code archiving method based on data cache under the cluster provided by the present invention reduces the amount of data transmitted by the network in the production cluster; The buffer is sent to the switch in the cluster, and then forwarded to the archiving node; and in the erasure code archiving method provided by the present invention, the data accessed by the user can be directly forwarded to the archiving node by using the switch in the cluster, avoiding the common correction In the code-deletion archiving method, the data is sent from the node buffer to the switch in the cluster network, which can effectively reduce the amount of data transmitted by the network in the production cluster and reduce the load of the entire network.

Description of drawings

Fig. 1 is a flow chart of the erasure code archiving method based on data cache under the cluster of the present invention;

Fig. 2 is a schematic diagram of an erasure code archiving method based on data caching under a cluster provided by an embodiment;

Figure 3 is a schematic diagram of user access and common erasure code archiving;

Fig. 4 is a schematic diagram of user access and data archiving based on data caching.

detailed description

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not constitute a conflict with each other.

For ease of understanding, the terms involved in the present invention are explained as follows:

Production cluster: refers to the general term for nodes that still need to provide data access services to upper-layer applications while providing archived data during the archiving process.

Erasure-coded archiving cluster: A cluster that stores complete data stripes after archiving is completed, which can be replaced by the free space of the production cluster itself.

Management server: manages the metadata information of the overall cluster, including the node location information of data blocks and their copies in the cluster.

Coding node: A node that implements coding calculation operations during the data archiving process of a stripe.

Node data blocks and stripes: In the encoding process of the archive, the unit of reading data is the data block. In a storage cluster, a stripe is composed of data block units with the same offset address on different nodes, which can independently recover the information set of invalid data.

Data bitmap: stored in the management server of the storage cluster, representing the correspondence between data blocks and data nodes.

NUM _j : During the archiving process, when the data blocks required for the current archiving stripe are not collected, the archiving node directly reads NUM _j = max{(μ-a _j -λ(j)) from the data node j in the production cluster ,0} data blocks, μ is the number of stripes that can be archived by the archiving node within time T, a _j is the number of data blocks that contain data node j in the archiving node, λ(j) means that in the next period of time T The number of data blocks received by the internal archiving node from node j.

Archived data status table: a global table maintained by the management server, recording whether the data block in the production cluster has been archived. In the embodiment, '0' and '1' are respectively used to indicate that the data block is 'not archived' and 'archived'.

The present invention provides an erasure code archiving method based on data cache under a cluster, which mainly uses cached data accessed by users to accelerate the erasure code archiving process. The specific process is shown in Figure 1, including the following steps:

(1) Read the required data blocks from the production node to the node buffer according to the user's access request, then send the data blocks to the access node, and forward it to the archive node according to the archive data status table; The steps specifically include the following sub-steps:

(1.1) The access node queries the management server according to the user request, and obtains the node position of the data block requested by the user in the production cluster;

(1.2) According to the position of the production cluster node obtained in step (1.1), the nodes of the production cluster read data blocks to the node buffer;

(1.3) The production node inquires the management server to obtain the archived data state table; according to the archived data state table, judge the archiving state of the user request data block;

(1.4) Data blocks in the production node sending buffer;

(1.5) The switch sends the data blocks to the access node to respond to user requests, and at the same time, forwards the unarchived data blocks to the archiving node to correspond to the archiving request;

(1.6) Judging whether all the processing of the user request is completed, if yes, the user visit ends; if not, then go to step (1.1), and continue to execute the next user request;

(2) The archiving node updates the number of data blocks of each data node according to the received user access data. If the archiving node receives all the data blocks of the current stripe, it will archive directly; if the data cannot be accessed through the user After receiving all the data blocks of the current stripe, the cold data is read from the production cluster and then archived; this step specifically includes the following sub-steps:

(2.1) The archiving node initializes the archiving stripe, the current archiving stripe number is i, i=0; the set a of the number of data blocks a _j including the data node j in the initializing archiving node is [0, 0, 0, 0 ];

(2.2) The archive node receives user access data (hot data) and follows the new collection a;

(2.3) The archiving node judges whether it has received all the data blocks of the archiving strip i, if so, enters step (2.7); if not, then enters step (2.4);

(2.4) The archiving node waits for a time interval T; within the time interval of the duration T, the archiving node continuously receives data blocks from the user cache; in this embodiment, T is 1 second;

(2.5) The archiving node follows the new set a, and judges again whether all the data blocks of the archiving stripe i have been received, if so, enter step (2.7); if not, enter step (2.6);

(2.6) The archive node directly reads cold data from the production cluster, and the number of data blocks read from data node j is Num _j , Num _j = max{(μ-a _j -λ(j)),0}, To get together μ strips, and then enter step (2.3);

(2.7) Archive the current stripe, and add 1 to the value of the stripe number i;

(2.8) update the archive data status table in the management server;

(2.9) Judging whether the archiving of all stripes in the production cluster has been completed, if not, proceed to step (2.3); if yes, end the archiving.

The erasure correction code archiving method based on data caching provided by the present invention is described below in conjunction with specific embodiments; what adopt in the example is (k+r, k) RS coding mode; Wherein, k refers to the number of original data blocks, and r represents The number of verification blocks generated by encoding, any k of k data blocks and r verification blocks can decode k original data blocks.

As shown in Figure 2, it is an example diagram of the erasure code archiving method based on data caching under the RS(6,4) encoding cluster in the embodiment; wherein, there are a total of six nodes in the erasure code archiving cluster, of which four It is a data node, and two are inspection nodes, using (6,4) RS coding method to realize and ensure data integrity, and the node numbers are {1,2,3,4,5,6};

Wherein, the logical address of any data block from the production cluster corresponds to one of the four data nodes in the erasure code cluster, and the archiving node archives the stripe number μ=3 within time T.

The erasure code archiving method based on data cache under the cluster provided in this embodiment is as follows:

(1) In the embodiment, the number of data blocks accessed by the user for the first time is [2, 4, 5, 3], that is, obtained from data node 1, data node 2, data node 3 and data node 4 respectively 2, 4, 5 and 3 data blocks;

The access node first queries the management server to obtain all the address information of the data block set; at the same time, the archive node initializes the stripe i; at this time, the number of data blocks of each data node in the archive node is [0, 0, 0 ,0]; among them, the initial value of i is 0;

(2) The production cluster reads the data block set to the buffer; when accessing for the first time, the switch forwards the data block set to the access node and the archive node at the same time;

(3) The access node obtains the data block set and responds to the user request; at the same time, the archiving node obtains the data block set, and updates the data block numbers of each data node to [2, 4, 5, 3];

The above steps (1)(2)(3) correspond to the process indicated by ① in Figure 2;

(4) The archiving node performs stripe i encoding calculation, sends 4 data blocks and 2 check blocks to the erasure code archiving cluster, i increases by 1, and i is 1 at this time; the archiving node clears the archived stripe 0 data blocks;

(5) If the archiving node receives all the data blocks of stripe i, it will archive; and send 4 data blocks and 2 check blocks to the erasure code archiving cluster, i increases by 1, at this time i is 2;

Delete the data block of stripe 1 that has been archived in the archive node; the number of data blocks of each node in the archive node is [0, 2, 3, 1];

The above steps (4) (5) correspond to the process indicated by ② in Fig. 2;

(6) If the archiving node has not received all the data blocks of the stripe i, it will enter the waiting state; at the same time, the user conducts the second visit, and the set of data blocks is [0, 3, 3, 4], assuming that the data If there is no data block duplicated with the first access in the block set, the switch forwards the data block set to the access node and the archiving node at the same time;

(7) The second visit of the corresponding user of the access node; at the same time, the archive node receives the data block set [0, 3, 3, 4], and the update node data block number is [0, 5, 6, 5];

Above-mentioned steps (6)(7) correspond to the flow process indicated by 3. in Fig. 2;

(8) After the archive node waits for the time period T, if it has not received all the data blocks of the stripe i, it will directly read the data blocks from the production cluster;

In the next time period T, the data block λ accessed by the user is [1, 1, 1, 1]. According to NUM _j = max{(μ-a _j -λ(j)), 0}, four The number of data blocks read by the data node is [2, 0, 0, 0]; two corresponding data blocks need to be read from data node 1; among them, the stripes that the archive node can archive within time T The number μ=3; the set a of the number of data blocks a _j of the data node j in the archiving node is [0, 5, 6, 5];

(9) The third user access data block set is [1, 1, 2, 1], the switch forwards the data block set to the access node and the archiving node at the same time; at the same time, the archiving node sets the data block set [2, 0, 0, 0] read data blocks directly from the production cluster;

(10) The archiving node receives the data block set [1, 1, 2, 1] from the user cache and the data block set [2, 0, 0, 0] read directly, and the number of data blocks to update the node is [3,6,8,6];

Above-mentioned steps (8)(9)(10) correspond to the flow process indicated by ④ in Fig. 2;

(11) The archiving node completes the archiving calculation of three stripes, and sends the data block and verification block of each stripe to the erasure code archiving cluster, and deletes the data block already archived in the archiving node; at this time i is 5, and the number of data blocks of each node in the archive node is [0, 3, 5, 3];

Above-mentioned step (11) corresponds to the flow process indicated by ⑤ in Fig. 2;

(12) Steps (1) to (11) are repeated until the archiving of all stripes in the production cluster is completed.

In the prior art, the general erasure code archiving process is shown in Figure 3, specifically: the archiving node directly reads the k data blocks required by the stripe from the production cluster, and then obtains the r collations of the stripe Finally, k data blocks and r verification blocks are sent to the erasure code archiving cluster for storage; each archive reads the data blocks from the production cluster and then calculates the verification blocks, and then Send data blocks and check blocks to the erasure code archiving cluster until the archiving is complete;

The erasure code archiving method based on data cache provided by the present invention is as follows:

(1) The archive node initializes the stripes and receives the data blocks of the user's access cache;

(2) If the k data blocks of the current archiving strip are all received, then enter step (4); otherwise, wait for the data sent from the data cache accessed by the user in the next time period T;

(3) Judging whether all the data blocks of the current stripe have been received, if so, enter step (4); if not, read the corresponding data blocks from the production cluster to get together μ stripes;

(4) The archiving node encodes and calculates the current strip, and sends k data blocks and r sum check blocks to the erasure code archiving cluster for storage;

(5) Steps (2) to (4) are repeated until the archiving of all stripes in the production cluster is completed.

In the embodiment, using the data block received from the user access cache in step (1), compared with the common erasure code archiving, greatly reduces the number of times the archiving node reads the data block from the production cluster, so that the The archiving operation is carried out without affecting the user access, and the archiving efficiency is improved; on the other hand, the data cached by the user is forwarded to the archiving node through the switch, which avoids the same effect on the production node as the user access and the ordinary erasure code archiving. Repeated read requests for data blocks can effectively reduce the amount of data transmitted by the network in the production cluster and reduce the load on the entire network.

Those skilled in the art can easily understand that the above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention, All should be included within the protection scope of the present invention.

Claims (3)

1. an erasure code archiving method based on data cache under a cluster, characterized in that, the erasure code archiving method specifically comprises the following steps: (1) According to the user's access request, read the required data blocks from the production node to the node buffer, and send the data blocks to the access node, and forward the data blocks to the archiving node according to the archive data status table ; (2) The archiving node updates the number of data blocks of each data node according to the received data blocks; and judges whether the archiving node has received all the data blocks of the current stripe, and if so, archives the current stripe ; If not, read the cold data block from the production cluster, and archive the stripes. 2. erasure code archiving method as claimed in claim 1, is characterized in that, described step (1) specifically comprises the following sub-steps: (1.1) The access node obtains the address of the required data block in the production cluster from the management server; (1.2) Send a read request to the node corresponding to the address of the data block in the generation cluster; (1.3) Read the data block from the production node in the generation cluster where it is located to the node buffer; (1.4) Judging whether the data block has been forwarded to the archiving node, if so, forwarding the data block to the access node; if not, forwarding the data block to both the access node and the archiving node. 3. The erasure code archiving method as claimed in claim 1 or 2, is characterized in that, described step (2) specifically comprises the following sub-steps: (2.1) The archiving node receives data blocks, and updates the number of data blocks of each data node included in the archiving nodes; (2.2) The archiving node initializes the current stripe, analyzes the node composition of the stripe, judges which data node the received data block belongs to, and initializes the storage variable; the node includes a data node and a check node; (2.3) Judging by the archiving node whether all the data blocks of the current stripe have been received, if so, enter step (2.7); if not, then enter step (2.4); (2.4) During the time interval of T, the archiving node receives data blocks from the user cache; (2.5) Judging whether the k data blocks of the current stripe are collected, if so, enter step (2.7); if not, then enter step (2.6); wherein, the k value is equal to the number of data nodes; (2.6) The archiving node directly reads the cold data from the production cluster to gather μ stripes, and then enters step (2.3); where μ is the maximum number of stripes that the archiving node can archive within the time interval T; (2.7) archive the current strip, and add 1 to the value of the strip number i, and update the archived data status table; (2.8) Judging whether the archiving of all the stripes is completed according to the archiving data status table, if not, then enter step (2.3); if yes, end the archiving.