CN106502828A - Optimized LVM-based remote copying method - Google Patents

Abstract

The invention particularly relates to an optimized LVM-based remote duplication method. The remote replication method based on LVM of this optimization, comprises LVM I/O Analysis module based on LVM remote replication framework, three parts of LVM I/O Process module and LVM Remote Replication module; Described LVM I/O Analysis module is applied to the I /O request analysis is optimized; the LVM I/O Process module changes the I/O processing flow of the write request; the LVM Remote Replication module changes the NBD module so that it is docked with the kernel structure of LVM to realize the remote data Send; the LVM Remote Replication module also optimizes the parameters of the TCP/IP network to make it more effective and suitable for remote replication scenarios. The optimized LVM-based remote replication method can be used in remote disaster recovery systems in the field of cloud computing. It does not require a special physical environment, and only needs to make corresponding improvements to the original IP network. Data is remotely replicated to another remote device, which improves the stability, security and transmission efficiency of data in remote replication.

Description

An Optimized Remote Copy Method Based on LVM

technical field

The invention relates to the technical field of cloud computing storage, in particular to an optimized LVM-based remote replication method.

Background technique

LVM (Logical Volume Manager, Logical Volume Management) is a management mechanism for dynamically managing storage devices in the Linux system. It provides the same continuous storage space as ordinary block storage to the upper layer by abstracting and encapsulating the underlying physical devices. LVM provides a complete user mode and kernel mode management mechanism. At the same time, the dynamic data migration, mirror, snapshot and other features provided by LVM technology can provide a complete framework for the data replication system.

NBD (Network Block Device) uses the disk space provided by the remote host locally as a block device by providing client and server services. NBD is a kernel module, which can be seamlessly connected with other block device managers (such as LVM) in the Linux system by defining the kernel interface.

The remote replication technology is to copy the data on the source device to one or more target devices through a remote connection. This technology is often used in disaster recovery systems. By utilizing geographical dispersion, when data in one area is lost, it can be quickly restored through remote data.

Based on the above situation, the present invention proposes an optimized LVM-based remote replication method.

Contents of the invention

In order to make up for the defects of the prior art, the present invention provides a simple and efficient optimized LVM-based remote replication method.

The present invention is achieved through the following technical solutions:

An optimized remote replication method based on LVM is characterized in that: the remote replication architecture based on LVM comprises LVM I/OAnalysis module, LVM I/O Process module and LVM Remote Replication module three parts; Described LVM I/O Analysis module is to The I/O request parsing of the upper-layer application is optimized, and the local data is directly accessed for the read request, and then sent to the local and remote device processing module for the write request; the LVM I/O Process module optimizes the Mirror interface in the LVM, and changes Write the requested I/O processing flow, by first recording the log and then updating the data, the data is sent to the local and remote devices respectively; the LVM Remote Replication module changes the NBD module so that it is docked with the kernel structure of the LVM, Realize the remote sending of data; Simultaneously, because NBD message is transmitted via TCP/IP, described LVM Remote Replication module has also been optimized to the parameter of TCP/IP network, adjusts tcp_sack in TCP/IP network, tcp_timestamps, tcp_low_latency, tcp_westwood and tcp_bic parameters to make it more effective for remote replication scenarios.

The LVM I/O Analysis module is responsible for filtering the I/O requests initiated by the upper-layer application, classifying the I/O requests by changing the lvm_make_request_fn() function, and directly handing over read requests or requests that have nothing to do with copying to the slave disk of LVM's own mechanism Read in, write request then call the interface of LVM I/O Process module to process; Described LVM I/OProcess module is responsible for the flow process of constructing I/O, docks with Remote Replication module simultaneously; Described LVM RemoteReplication module is responsible for remote replication Data consistency definition and remote transmission of data; after the I/O request is processed by the LVM Remote Replication module on the master end, it is sent to the remote device end and the log is successfully submitted and written to the data disk, and the local log information is updated and submitted to LVM I/O Process module; when the mirror in the LVM I/O Process module receives the local and remote return messages, it returns the write success to the application layer.

The I/O processing flow of the write request by the LVM I/O Process module first records the I/O request into the log information, and then copies the BIOS in two copies, one is sent to the local data disk, and the other is sent to the local data disk. Into the request queue, finally call the Remote Replication module interface, and use multi-threaded processing to send the data to the peer end; when the two requests from the local and remote devices are received, the request processing success is returned to the upper layer application.

The LVM Remote Replication module defines the data consistency of remote replication as adopting the method of real-time synchronization, that is, multiple write requests sent by the upper layer application must wait for the data of the master end and the backup end to be written simultaneously before continuing to execute; when the request arrives When writing logs, first submit the logs in the order of requests, and then write to the data disk after writing the logs successfully. When writing to the data disks, write to the disks in sequence. Finally, notify the upper layer that the request is complete;

The remote transmission of data by the LVM Remote Replication module is based on the NBD module. First, the primary end and the standby end are initialized and connected through the nbd_init function, and then the do_nbd_request function is used to read the corresponding request from the request queue, using the message structure The handle field in is loaded and sent to the standby; call nbd_do_it to wait in a loop and call nbd_read_stat to read the write status of the standby; when the read and write request is completed, delete the request from the request queue and return success to the upper layer.

The optimization of the parameters of the TCP/IP network by the LVM Remote Replication module includes using the tcp_sack parameter to enable a selective response mode, improving the reply of out-of-order messages, and only sending lost messages; using the tcp_timestamps parameter to enable TCP timestamps, which is more accurate Control packet retransmission; turn off the tcp_low_latency parameter and select the tcp_westwood parameter to optimize the overall utilization of the bandwidth through throughput evaluation; enable the Binary Increase Congestion of the tcp_bic parameter for fast and long-distance transmission to make better use of the GB-level speed of the transmission network.

The beneficial effects of the present invention are: the optimized LVM-based remote replication method can be used in the remote disaster recovery system in the field of cloud computing, and does not require a special physical environment, and only needs to make corresponding improvements to the original IP network , Remotely copy data to another remote device with a complete software method, which improves the stability, security and transmission efficiency of data in remote copying.

Description of drawings

Accompanying drawing 1 is the structural diagram of the present invention based on LVM remote replication architecture.

Accompanying drawing 2 is a schematic diagram of the processing flow of the LVM I/O Process module of the present invention.

Accompanying drawing 3 is the schematic diagram of IO flow optimization of LVM Remote Replication module of the present invention.

Accompanying drawing 4 is a schematic diagram of the processing flow of the LVM Remote Replication module of the present invention.

Accompanying drawing 5 is a schematic diagram of the message structure of the NBD submodule in the LVM Remote Replication module of the present invention.

detailed description

In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention clearer, the present invention will be described in detail below in conjunction with the accompanying drawings and embodiments. It should be noted that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

The remote replication method based on LVM of this optimization, comprises LVM I/O Analysis module based on LVM remote replication architecture, LVM I/O Process module and LVM Remote Replication module three parts; Described LVM I/OAnalysis module is applied to the I/O O request analysis is optimized, directly accesses local data for read request, then sends to local and remote device processing module for write request; Described LVM I/O Process module optimizes the Mirror interface in LVM, changes the I/O of write request 0 processing flow, by first recording the log and then updating the data, the data is sent to the local and remote devices respectively; the LVM Remote Replication module changes the NBD module so that it is docked with the kernel structure of the LVM, and realizes the remote sending of the data Simultaneously, because NBD message is transmitted via TCP/IP, described LVMRemote Replication module also optimizes the parameter of TCP/IP network, adjusts tcp_sack in TCP/IP network, tcp_timestamps, tcp_low_latency, tcp_westwood and tcp_bic parameter , making it more effective for remote replication scenarios.

The LVM I/O Analysis module is responsible for filtering the I/O request initiated by the upper layer application, classifying the I/O request by changing the lvm_make_request_fn () function, and the main function of the lvm_make_request_fn () function is to map the Mapped Device to the The underlying Target Device, and returns the mapped address to the system, read requests or requests not related to copying are directly handed over to LVM’s own mechanism to read from the disk, and write requests are processed by calling the interface of the LVM I/OProcess module; The LVM I/O Process module is responsible for building the I/O process and is connected to the Remote Replication module. The entire LVM I/O Process module is implemented based on the mirror feature change of LVM. The mirror feature sends I/O requests to the local respectively. and in the replication volume; the LVM Remote Replication module is responsible for the definition of data consistency of remote replication and the remote transmission of data; after the I/O request is processed by the LVMRemote Replication module at the master end, it is sent to the remote device end and the log write is successfully submitted After arriving at the data disk, update the local log information and submit it to the LVM I/O Process module; when the mirror in the LVM I/O Process module receives the local and remote return messages, it returns the write success to the application layer.

The I/O processing flow of the write request by the LVM I/O Process module first records the I/O request into the log information, and then copies the BIOS in two copies, one is sent to the local data disk, and the other is sent to the local data disk. Into the request queue, finally call the Remote Replication module interface, and use multi-threaded processing to send the data to the peer end; when the two requests from the local and remote devices are received, the request processing success is returned to the upper layer application.

The request queue is a buffer for I/O requests. This mechanism is introduced because the transmission speed of the remote replication I/O that needs to be sent to the remote device is very different from the local storage speed. In order to improve the overall I/O writing performance, it is considered to store the I/O request in the queue and use Multi-threaded processing sends data to the peer, thereby optimizing the overall write request processing of the system.

The LVM Remote Replication module defines the data consistency of remote replication as adopting the method of real-time synchronization, that is, multiple write requests sent by the upper layer application must wait for the data of the master end and the backup end to be written simultaneously before continuing to execute; when the request arrives When writing logs, first submit the logs in the order of requests, and then write to the data disk after writing the logs successfully. When writing to the data disks, write to the disks in sequence. Finally, notify the upper layer that the request is complete;

The remote transmission of data by the LVM Remote Replication module is based on the NBD module. First, the primary end and the standby end are initialized and connected through the nbd_init function, and then the do_nbd_request function is used to read the corresponding request from the request queue, using the message structure The handle field in is loaded and sent to the standby; call nbd_do_it to wait in a loop and call nbd_read_stat to read the write status of the standby; when the read and write request is completed, delete the request from the request queue and return success to the upper layer.

The optimization of the parameters of the TCP/IP network by the LVM Remote Replication module includes using the tcp_sack parameter to enable a selective response mode, improving the reply of out-of-order messages, and only sending lost messages; using the tcp_timestamps parameter to enable TCP timestamps, which is more accurate Control packet retransmission; turn off the tcp_low_latency parameter and select the tcp_westwood parameter to optimize the overall utilization of the bandwidth through throughput evaluation; enable the Binary Increase Congestion of the tcp_bic parameter for fast and long-distance transmission to make better use of the GB-level speed of the transmission network.

Claims (5)

1. an optimized LVM-based remote replication method, characterized in that: the LVM-based remote replication architecture includes LVMI/O Analysis module, LVM I/O Process module and LVM Remote Replication module three parts; said LVM I/O Analysis The module optimizes the analysis of the I/O request of the upper layer application, directly accesses the local data for the read request, and sends the write request to the local and remote device processing module; the LVM I/O Process module optimizes the Mirror interface in the LVM , change the I/O processing flow of the write request, and send the data to the local and remote devices respectively by first recording the log and then updating the data; the LVM Remote Replication module changes the NBD module so that it is consistent with the kernel structure of the LVM docking to realize remote sending of data; at the same time, since the NBD message is transmitted via TCP/IP, the LVM Remote Replication module also optimizes the parameters of the TCP/IP network, and adjusts tcp_sack and tcp_timestamps in the TCP/IP network , tcp_low_latency, tcp_westwood and tcp_bic parameters to make it more effective for remote replication scenarios. 2. the remote copy method based on LVM of optimization according to claim 1, is characterized in that: described LVM I/OAnalysis module is responsible for the I/O request filtering that upper layer application initiates, by changing lvm_make_request_fn () function to I/O O requests are classified, read requests or requests that have nothing to do with copying are directly handed over to LVM’s own mechanism to read from the disk, and write requests are processed by calling the interface of the LVM I/O Process module; the LVM I/O Process module is responsible for constructing The process of I/O is docked with the Remote Replication module at the same time; the LVM Remote Replication module is responsible for the data consistency definition of remote replication and the remote transmission of data; after the I/O request is processed by the LVM RemoteReplication module at the master end, it is sent to After the remote device successfully submits the log and writes it to the data disk, it updates the local log information and submits it to the LVM I/O Process module; when the mirror in the LVM I/O Process module receives the local and remote return messages, it sends a message to the application The layer returns write success. 3. the optimized LVM-based remote copy method according to claim 1 or 2, characterized in that: the LVM I/O Process module first records the I/O request to the I/O processing flow of the write request In the log information, the BIOS is copied twice, one is sent to the local data disk, the other is sent to the request queue, and finally the RemoteReplication module interface is called to send the data to the peer end by using multi-thread processing; when the local After the two requests of the remote device and the remote device are completed, the request processing success is returned to the upper layer application. 4. the LVM-based remote replication method of optimization according to claim 2 is characterized in that: the data consistency definition of the remote replication by the LVMRemote Replication module refers to the mode of adopting real-time synchronization, that is, multiple The write request must wait for the data on the primary end and the standby end to be written at the same time before continuing to execute; when the request arrives, the log is first submitted in the order of the request, and the data disk is written after the log is successfully written. When the data disk is written, the disk is written in sequence. Only after writing the log and writing the data disk are successfully submitted at the same time, the log information is updated in the order of request submission, and finally the upper layer is notified that the request is completed; The remote transmission of data by the LVM Remote Replication module is based on the NBD module. First, the primary end and the standby end are initialized and connected through the nbd_init function, and then the do_nbd_request function is used to read the corresponding request from the request queue, using the message structure The handle field in is loaded and sent to the standby; call nbd_do_it to wait in a loop and call nbd_read_stat to read the write status of the standby; when the read and write request is completed, delete the request from the request queue and return success to the upper layer. 5. the LVM-based remote replication method of optimization according to claim 1 is characterized in that: the optimization of the parameters of the TCP/IP network by the LVMRemote Replication module includes utilizing the tcp_sack parameter to enable a selective response mode, and improving out-of-sequence reporting Reply to text, only send lost packets; use tcp_timestamps parameter to enable TCP timestamp to control packet retransmission more precisely; disable tcp_low_latency parameter and select tcp_westwood parameter to optimize the overall utilization of bandwidth through throughput evaluation; The distance transmission enables the Binary Increase Congestion of the tcp_bic parameter to make better use of the GB-level speed of the transmission network.