Cache synchronization method and equipment for distributed cluster file system

Abstract

The embodiment of the invention provides a cache synchronization method and equipment for a distributed cluster file system. According to the method, a global cache table is defined for an object file and includes the record that each node modifies the object file; when a first node performs write addressing on the object file, a latest global cache table is obtained according to the information of a second node which modifies the object file lastly, wherein the information of the second node is included in a writing lock, and the data that the second node modifies the object file is obtained according to the record that the second node modifies the object file lastly, wherein the record is included in the latest global cache table; when the two nodes write different data of the object file interactively at the same time, the data in caches of the two nodes are valid, and follow-up reading can be obtained from the caches; when the two nodes write same data interactively, modification data is needed to be obtained from another node. The efficiency that data is transmitted in a network is far higher than that of once disk writing and disk reading, so that the systematicness of cluster files is greatly improved when compared with that of a conventional manner.

Description

technical field [0001] Embodiments of the present invention relate to data communication technology, and in particular to a cache synchronization method and device for a distributed cluster file system. Background technique [0002] The distributed cluster file system can solve the problem of massive data storage. In the cluster, multiple devices are combined to provide external services, and multiple nodes (servers) are used to replace one node to complete the task. This not only improves the processing capacity, but also ensures high reliability. The symmetrically distributed cluster file system has a globally unified namespace. The storage space is visible and identical to each node. Metadata and data can be distributed to any node in the cluster. Of course, the system can also be accessed from any node. All data stored. [0003] Current storage systems are divided into at least two levels of storage: storage physical media and caches created from main memory. In order...

AI Technical Summary

Problems solved by technology

After node 2 writes access to file 1, it needs to modify the fourth data block of file 1. Distributed lock finds that node 1 has applied for a write lock, so it reclaims the write lock of file 1 from node 1, and node 1 receives the command to reclaim the write lock After that, the cached data of file 1 is first downloaded to the disk. Since node 2 applies for a write lock, it means that node 2 wants to modify the data of file 1, so the cached data related to file 1 on node 1 may not be the latest , it needs to be invalidated, and then return the lock authority to the lock management device. The lock management device authorizes the lock authority to node 2. After node 2 obtains the write lock, it modifies the corresponding data to the cache. After the modification, node 1 reads and accesses file 1 The 0th data block of , as before, the lock management device reclaims the write lock from node 2. During the recovery process, the data of node 2 is written to the disk. Since node 1 applies for a read lock, reading will not modify the data, so the cache of node 2 It is still valid, that is, there is no need to invalidate the corresponding cache, and then the lock management device authorizes the read lock to node 1, and node 1 finds the corresponding data from the cache, but finds that the data is invalid, indicating that the data may not be up-to-date. disk read

[0005] From the above example, we can see that by writing the latest data to the disk to ensure the consistency of the data accessed by each node, there will be many disadvantages as follows: If the nodes in the cluster read and write the same file interactively, this will cause File data is frequently read and written to the disk, the cache does not have any acceleration effect, but increases the complexity of the system, and frequent disk reading and writing will increase the delay of the system, which will greatly affect the performance

Images