Detailed description of VMware DRS resource scheduling technology. Doc

VMware distributed resource scheduling (DRS) continuously monitors the utilization of resource pools in VMware Host clusters, it can intelligently allocate resources in virtual machines based on business needs. By dynamically allocating and balancing computing resources, VMWare DRS can integrate servers to reduce it costs, increase flexibility, and reduce downtime to ensure business continuity and stability; reduce the number of servers to be run and dynamically cut off the power of unused servers, improving energy utilization.

Once the customer integrates the server to a physical host with fewer resources, the resource requirements of virtual machines often become unexpected bottlenecks, and the demand for all resources may exceed the available resources of the host. VMware DRS provides an automation mechanism to migrate virtual machines to hosts with more available resources by continuously balancing capacity, ensure that each virtual machine can call corresponding resources in a timely manner on any node. Even if a large number of SQL Server virtual machines are running, as long as DRS is used, the CPU and memory bottlenecks do not need to be monitored one by one. Fully Automated resource allocation and load balancing functions can also significantly reduce the cost and operation costs of data centers.

VMware DrsOperation Mode

VMware DRS allows users to define their own rules and solutions to determine the way virtual machines share resources and determine their priorities. When the workload of a virtual machine increases, VMWare DRS evaluates the priority of the Virtual Machine Based on the previously defined allocation rules. If the Virtual Machine passes the evaluation, DRS allocates additional resources for it. When the host resources are insufficient, DRS will look for hosts with redundant resources in the cluster, migrate the VM to call more resources for its heavy load service.

VMware DRS automatically and intelligently balances resources in virtual machines

There are two ways for DRS to allocate resources: migrating a virtual machine to another server with more suitable resources, or migrating other virtual machines on the server, this will free up more "space" for the virtual machine ". Virtual Machine real-time migration on different physical servers is implemented by vmotion. VMware DRS supports both automatic and manual modes. In the automatic mode, DRS determines the optimal allocation scheme of virtual machines between physical servers, and automatically migrates virtual machines to the most suitable physical servers. In manual mode, VMWare DRS provides an optimal solution for placing virtual machines, and the system administrator determines whether to adjust the virtual machines according to the solution.

Drs resource allocation method 1

Drs resource allocation method 2

The flexible hierarchical layout of resource pools ensures that available IT resources are synchronized with the requirements of commercial organizations. Business departments can not only obtain dedicated IT resources, but also enjoy all the benefits of resource pools. This maximizes the utilization of all performance resources and reduces unnecessary expenditures and consumption.

Deploy VMware Drs

Deploying VMware DRS is actually very simple. The complex one is mainly to develop the DRS thresholds and rules, and provide each host with the shared processor performance and memory capacity, the so-called threshold value is a system water level threshold, which can be customized according to the hardware configuration of the physical machine. When the resource usage reaches this threshold value, VMware's cluster manager automatically performs operations based on rules or provides administrators with an operation suggestion. Compared with clusters in the traditional sense, VM clusters based on Vmware DRS can better reflect smart and automated configuration technologies.

You can configure DRS and HA when deploying a cluster.

When you deploy a cluster, you can configure DRS and HA together or separately, and decide whether to perform ha high availability and dynamically allocate resources for each virtual server in the cluster.

Select various automation levels based on the complexity of clusters and Applications

VMware DRS can select multiple levels of automation policies, either manually or extremely "radical" automatic DRS policies. If it is a fully automated policy, the Cluster Manager will migrate the server according to the rules and thresholds. This process may not be prompted. During the test, we often find that, the virtual server has actually switched the host location, but this was an "event later" Discovery.

Power Management DPM of VMware DRS has the same options as automatic management. When the power management function is enabled, resources of multiple hosts are combined for computing, the manager will automatically centralize multiple servers on one host, and then the redundant hosts will enter sleep to save energy. When the demand for resources increases, it will automatically wake up the sleeping host to provide more available resources.

Power Management for VMware Drs

VMware EVC

VMware EVC (enhanced vmotion compatibility) is an online hot migration solution supported by the underlying CPU hardware (implemented by Intel through FLEX migration, and implemented by AMD through extended migration ), the cpuid mask function is supported on the Virtual Layer. Some CPU functions are shielded and vmotion between different CPU generations is achieved by force CPU Instruction Set compatibility.

Vmotion is very important in practice. However, vmotion has strict CPU requirements and cannot automatically implement vmotion for different series of CPUs. During the implementation of VI, the server CPU in A group is different. For example, the old server is based on Intel 5300 CPU while the new server is based on Intel 5500 CPU. Because the new processor adds a new SSE instruction set, after the new server is added to the cluster, the virtual machine cannot perform vmotion operations from the old server to the new server.

Before EVC is available, the solution is to manually calculate the CPU mask, modify the configuration file of each virtual machine one by one, or globally modify vpxd. cfg on the VC server to achieve compatibility between different series of CPUs. The disadvantage of this solution is that you need to manually design the CPU masking compatible with all servers in the cluster. When a new server is added to the cluster, in order to avoid possible incompatibility, you need to manually calculate a new CPU masking again.

After the EVC function is used, all the above manual work is completed by VC and ESX Server. VMWare works with the CPU manufacturer to directly support the cpuid mask function at the underlying hardware layer. However, even if you use EVC,Nor can we migrate virtual machines on Intel servers to AMD hosts.. Therefore, EVC has only three options:Disable, Intel, or AMD.

Cluster configuration complete

Add a new VM on the host to automatically add the newly created Cluster

After this cluster is configured, the VM above can use the features of VMware DRS. At the cluster or virtual machine level, you can specify the operation mode of DRS: automatic, manual, and semi-automatic. VMware recommends that automatic mode be used by default for clusters. Manual mode is recommended for virtual machines that are sensitive to vmotion. In this mode, the administrator can decide whether or not to migrate the VM. For general virtual machines, the automatic mode should be used as much as possible. In the automatic mode, virtual machines can be automatically migrated according to the balance of workload, thus improving the efficiency.

VsphereThere are many new features in DRS, the most significant of which are the following two points:

1. Fully support the new distributed power management (DPM ). This feature is experimental in earlier versions of vi3 and is fully available in vsphere.

VMware DPM reduces energy consumption by balancing the workload of the data center. As part of VMware DRS, the distributed power management (DPM) will automatically cut off the power of servers that are not currently needed and restart these servers as computing resources demand increases.

When the cluster utilization is low, the DRS with dpm function will reintegrate the workload to a few ESX server hosts, and we recommend that you power off some hosts that are temporarily not needed. When the workload of the cluster increases and more resources are required, DRS with dpm function will recommend that you power some ESX server hosts again, and balance the workload of hosts in the cluster again. DPM also ensures that the capacity of all power supply hosts meets the VMware ha setting requirements.

Like DRS, DPM has two operation modes: automatic and manual. The VMware DRS cluster can use the DPM feature on any ESX server host as long as the hardware device supports DPM with the appropriate configuration. For example, the NIC that uses the vmkernel network must have the wake-on-lan function. This function is mainly used to start the ESX server host from the power-off status.

2. Another key feature is vapp groups. Vapp groups are used to store virtual machines and manage them as an independent whole, including resource management.

VAPPS compress multi-layer applications to a vapp entity to simplify the deployment and subsequent management of multi-layer applications. VAPPS not only compresses virtual machines, but also dependencies and resource allocation between them, allowing you to perform one-step power supply operations, replication, deployment, and monitoring on the entire application.

After the data center is successfully created and a host cluster with DRS function has been added to the vcenter server system, you can create a vapp (the cluster with DRS function is selected in the directory ). It is worth noting that when a DRS cluster starts a vapp in manual mode, DRS does not generate a recommended Virtual Machine Deployment Solution. The VM allocation method for this startup operation is as if Drs is running in Semi-automatic or automatic mode. However, this does not affect vmotion suggestions.