Shulou(Shulou.com)06/02 Report--

Not long ago, a netizen asked Microsoft under Lao Wang's blog about Microsoft's dual-machine dual-cabinet solution, which has the highest performance-to-price ratio and requires the least hardware investment. Lao Wang found this topic very interesting and specially shared the discussion with you today.

According to Lao Wang's understanding, the so-called dual machine and dual cabinet refers to two servers, each connected to its own storage cabinet, and then the two servers are replicated to ensure storage support in case of application failover.

According to this concept, Microsoft has two dual-machine and dual-cabinet solutions, one is Hyper-V replication, the other is storage copy, let's analyze them one by one.

Hyper-V replication is a virtual machine replication technology introduced by Microsoft in 2012 hyper-v3.0, which supports replication of virtual machines to other hosts or clusters or Azure, and disaster recovery of virtual machines without clustering.

Advantages and disadvantages of Hyper-V replication technology

Storage independence. Hyper-v hosts participating in replication can connect to different storage and can use existing enclosures or local disks

Support for standard and data center editions

Support workgroup deployment

The replication process uses port 80 or 443 and supports certificate encryption

Support stand-alone replication, cluster-to-cluster replication, stand-alone replication, cluster-to-stand replication

Support for extended replication, Amurb BMI C (introduced by 2012R2)

Support for planned failover

Disaster recovery is supported, but manual failover is required at the backup node

Support multi-subnet architecture TCP/IP presets. For example, you can set the virtual machine to be automatically set to slave site IP when the virtual machine is transferred to the slave site.

Support for reverse replication

Support for LAN or WAN environments

The replication process distinguishes the active and standby architectures. for a virtual machine participating in replication, only the primary node starts up, the standby node shuts down, and the primary node replicates the virtual machine incremental data to the standby node according to the time interval.

Support test failover, set up a test virtual machine, and monitor whether the data is copied properly

Support for replication of virtual machines to Azure through ASR

Hyper-V virtual machine level replication, which supports replication of all virtual disks of virtual machines (except pass-through)

Support for multiple recovery points

Support application consistency. If the virtual machine assumes database services such as SQL, and replicates the virtual machine with application consistency, you can create recovery points after the in-memory data is Flush to disk during replication.

With official replication planning software, Microsoft SCOM platform can monitor it, but SCVMM cannot operate hyper-v replication.

To sum up, hyper-v replication can be said to be a disaster recovery solution for virtual machine replication. Lao Wang thinks that the hardware change is the smallest of several solutions. You can directly use the existing dual-machine, double-cabinet architecture. There is no requirement for storage. Replication itself does not take up too much system resources. Various replication functions, such as extended replication, test failover, recovery control, architecture support, application awareness. Data security is relatively perfect, but the main disadvantage is that unplanned failures must be failed over manually, and the native cannot fail over automatically. If you need automatic failover, you also need to learn SCO or use ASR, which is only suitable for Hyper-V virtual machine load, but not if virtualization is not used.

Storage replica technology is a block-based replication technology introduced by Microsoft Windows Server 2016, which replicates partitions at the system level, supports partition-to-partition, stand-alone, extended cluster, cross-cluster replication and other disaster recovery scenarios to help organizations improve business continuity.

Summary of advantages and disadvantages

Only 2016 data Center Edition is supported

Nodes participating in storage replication must join the domain

System-level partition replication, and the application is not aware of the underlying replication

Replication uses SMB 3.1.1 communication protocol, port 445

Support for synchronous and asynchronous replication

Storage has no correlation, and the bottom layer of the node can be any storage structure

Support for fixed disks and thin standby disks

There is an active / standby relationship in the replication process. The primary replication partition can be read and written, while the standby replication partition cannot be read or written. Standby read-only mode is not supported.

For stand-alone and cross-cluster replication, you can only use Powershell commands to manage replication, and extend cluster support for GUI management.

Manual failover is required for stand-alone and cross-cluster replication, extending the cluster to achieve fully automatic failover

The extended cluster still requires each site to access its own site storage, and the direct use of each node's local disk is not supported.

Cross-cluster replication supports two clusters using different architectures for storage

Log disk and data disk are required for replication. The data is first written to the log disk, and then Commit data disk.

The replication node requires at least two disks, one data disk and one log disk

The format of data disk and log disk must be GPT. MBR format disk is not supported.

Both data disks and partitions must be the same size, maximum 10TB

Both log disks and partitions must be the same size, at least 8GB

Does not support extended replication, multiple recovery points, application consistency awareness, workgroups, cluster-to-stand-alone

Support through SCOM monitoring, SCVMM management, OMS monitoring, Honolulu management

To sum up, storage replication is a system-level block storage replication above server 2016, which works under the volume above the partition. For standalone-to-stand-alone, the storage bottom layer does not have many restrictions, but for extended clusters, each node is still required to connect to its own storage cabinets. The difference between storage replication and Hyper-V replication is that storage replication is only implemented at the system level and is not bound to virtualization. Therefore, most applications can make use of this function, and the disadvantage is that although storage replication is also a disaster recovery technology, stand-alone to stand-alone can achieve storage replication, manual switching, and extended cluster can achieve fully automatic failover of storage plus computing, but from the perspective of disaster recovery software, storage replication still lacks some functions, such as extended replication function is not supported. The cluster does not support stand-alone functions, resulting in a very inflexible architecture for storage replication, which also needs to be improved in the future. Different from Microsoft's DFS, storage replication is not as complex as DFS principle, detecting NTFS USN, updating DFSR DBID, updating GVSN,RDC synchronous data, storage replication uses log + storage mechanism, mainly detects disk write IO, detects write IO Log and storage writes are performed according to synchronous replication mechanism or asynchronous replication mechanism, so storage replication can copy open files. When discussing usage scenarios, Lao Wang thinks that stand-alone replication is suitable for the following scenarios.

1. Files are often opened and sometimes will not be copied using DFS, so you can use storage to copy the whole partition

two。 For VHDX files, VMM library sharing, database MDF result set files, such large files, which are often opened, can be copied directly on two nodes.

3. It is suitable for archiving scenarios, for example, some virtual machines are shut down all the year round, but there is some key data in it. At this time, you can use storage replication to simply copy the virtual disk of the archived virtual machine.

The above is the stand-alone scenario of storage replication envisaged by Lao Wang. Another point is to extend the cluster, which can be said to be the biggest highlight of storage replication, that is, using asymmetric storage architecture and mounting storage on their respective nodes. through the cooperation of storage replica and cluster, we can fail over storage first and then fail over the upper virtual machine in the event of failure.

If you see a lot of documents, you may say that the extension cluster is suitable for multi-site architecture. for example, the respective sites of Beijing and Shanghai sites only access simple storage cabinets and are not configured with hardware-level storage replication. but through the Microsoft 2016 extension cluster, when the Beijing station is down, it can be started in Shanghai together with storage and computing resources, which is indeed the most important significance of extending the cluster. Help the company save a lot of hardware costs, two points, one is to achieve native storage replication, one is to achieve storage replication and cluster convergence, cluster detection node downtime, automatic failover storage and applications.

However, this kind of automatic failover storage and application cannot be realized without multiple sites. The answer is no, even if you only have one data center, one computer room, two nodes, two storage cabinets, each node is connected to the cabinet, you can extend the cluster architecture, as long as it meets the configuration requirements of the extended cluster. The extended cluster does not care whether you are a local data center or a remote data center, depending on your technology selection

The benefits of extending the cluster architecture are the convergence of storage replication and clustering, perfectly handling a single point of failure of storage and a single point of failure of an application, and achieving storage and cluster node failover linkage, so if you want this feature in your local data center, it is also OK.

Storage replication vs. Hyper-V replication

Storage replication is system specific, replication of system disks is not supported, Hyper-V replication is virtual machine level, replication of pass-through type disks is not supported

Storage replication can be used for physical machines, virtual machines, private clouds and public clouds, as long as there is OS, and hyper-v replication needs to be configured on physical machines.

Neither of them requires additional storage replacement, and both can use existing storage cabinets to achieve obsolescence.

Storage replication is to copy the storage, do a good storage switch, hyper-v replication will contain the memory state, but also copy the memory state of the master node.

Storage replication is only available in the 2016 data center version. If the enterprise does not need additional construction, and 2016 requires more than 2012 for the system, if you configure the storage replication function, it is recommended to reserve at least 2-4GB memory for the system.

Storage replication is implemented as a log disk plus data disk architecture. Administrators need to manage log disks and data disks, while hyper-v replication does not have this architecture.

Storage replication is not based on checkpoints, but on continuous replication, so the increment of change is often much lower than that of snapshot-based products.

If you want to replicate an application at the virtual machine level, and you want to enable the virtual machine to boot directly, and the application transaction is not lost, it is recommended to use Hyper-v replication with application consistency.

If you want to copy VHDX libraries, database MDF result set files, archive virtual machines, or some daily files that will be opened, you can use storage replication

If you want to achieve dual-machine, dual-cabinet storage replication + automatic failover, it is recommended to use an extended cluster

If you want a more flexible replication architecture such as extended replication at the virtual machine level, Hyper-V replication is recommended

The above is Lao Wang's talk about Microsoft's serious dual-machine and dual-cabinet solution, hyper-v replication and storage replication. the main advantages of both are that they can directly benefit from the old, using existing storage cabinets and without changing the infrastructure can achieve disaster recovery, hyper-v replication can not achieve automatic failover, storage replication can be configured to extend the cluster to achieve automatic failover, this may be a major point of consideration. In addition, storage replication takes up more system CPU and memory resources than hyper-v replication. Hyper-v replication is more professional in disaster recovery than storage replication, and is more suitable for overall replication of important virtual machines. When you actually use it, you can choose the applicable scenario combined with these function points mentioned by Lao Wang.

In addition to these two schemes that can directly benefit from the old, Microsoft 2016 also introduces the function of S2D for such scenarios where there is no shared storage, which extends the storage pool and storage space functions of 2012R2 to multiple nodes, which is implemented in the following two points

The local disks on each cluster node can be grouped together, each node will have a clusport component acting as an adapter card, and there will be a clusblft role in the cluster, which is responsible for connecting the clusport of each node. Through these two components, the local disks that meet the requirements on each cluster node can be gathered together to form a logical storage pool of the cluster, which can further configure the storage space. But in fact, this logical storage pool is the local disk of each cluster node. After the storage space is built, the generated virtual disk can be used for the cluster disk.

Another point is the new fault-tolerant mechanism implemented after the storage space is extended to the cluster, 2016 S2D. According to the different fault-tolerant configuration of the storage space, the data write can be divided into multiple extent, one extent 1GB. If configured as a double mirror, the data background will generate two extent, and the two extent can be sprinkled on different nodes, different cabinets, and different racks. Through this function, the storage space can be extended to the cluster. The failure of one node or disk will not affect the reading of the upper application, because all the disks will be read on the other node, and the data will be rebuilt on the other node. S2D will follow the principle of extent fault tolerance and anti-correlation, and always scatter multiple extent of the same data on different nodes. When an IO comes in, the IO will end only when all the extent of the data has been written.

Therefore, if a newly deployed environment, it does not matter whether it is good or bad, then you can try this new super-fusion architecture, which is not used in asymmetric storage architecture. Direct super-fusion uses local storage of each node, and data is always fault-tolerant and written to different nodes to ensure that the data is highly available, and supports SSD,NVME,NVDIMM-N cache settings, which can achieve higher performance. S2D consumes more resources than storage replication, especially network and memory, and S2D is not a backup scheme or disaster recovery solution. Once the function of S2D is not work, data reading will be very troublesome, so if you use S2D, you need to consider a separate backup mechanism.

In addition to Microsoft's own three solutions, DataKeeper SIOS,Starwind,Symantec SFW is also a good solution. SIOS supports the replication of local storage or asymmetric cabinets on each node, and the replication configuration of storage can be completed on SIOS. The disk replicated by SIOS can be directly displayed on the WSFC cluster disk. It is also good if there is no server2016 to use this tool. It is perfectly integrated with the cluster and can also achieve automatic failover. SFW is similar to it. It manages the storage of different nodes, builds disk groups, and then delivers them to the cluster. Starwind connects the storage of each node, replicates, replicates the virtual disk, and then provides it to each cluster node in the way of ISCSI.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.