VMware vSphere® Distributed Resource Scheduler™ (DRS) is more than a decade old and is constantly innovating with every new version. In vSphere 6.5, DRS comes with many new features and performance improvements to ensure more efficient load balancing and VM placement, faster response times, and simplified cluster management.
In this paper, we cover some of the key features and performance improvements to highlight the more efficient, faster, and lighter DRS in vSphere 6.5.
Historically, vSphere DRS has been reactive—it reacts to any changes in VM workloads and migrates the VMs to distribute load across different hosts. In vSphere 6.5, with VMware vCenter Server® working together with VMware vRealize® Operations™ (vROps), DRS can act upon predicted future changes in workloads. This helps DRS migrate VMs proactively and makes room in the cluster to accommodate future workload demand.
For example, if your VMs’ workload is going to spike at 9am every day, predictive DRS will be able to detect this pattern before-hand based on historical data from vROPs, and can prepare the cluster resources by using either of the following techniques:
- Migrating the VMs to different hosts to accommodate the future workload and avoid host over-commitment
- Bringing back a new host from stand-by mode using VMware vSphere® Distributed Power Management™ (DPM) to accommodate the future demand
How It Works
To enable predictive DRS, you need to link vCenter Server to a vROps instance (that supports predictive DRS), which monitors the resource usage pattern of VMs and generates predictions. Once vROps starts monitoring VM workloads, it generates predictions after a specified learning period. The generated predictions are then provided to vCenter Server for DRS to consume.
Once the VMs’ workload predictions are available, DRS evaluates the demand of a VM based on its current resource usage and predicted future resource usage.
- Demand of a VM = Max (current usage, predicted future usage)
Considering the maximum of current and future resource usage ensures that DRS does not clip any VM’s current demand in favor of its future demand. For the VMs which do not have predictions, DRS computes resource
demand based on only the current resource usage.
Look Ahead Interval
The predictions that DRS gets from vROps are always for a certain period of time, starting from the current
time. This period is known as the “look ahead interval” for predictive DRS. This is by default 60 minutes starting from the current time, which means, by default the predictions will always be for the next one hour. So if there is any sudden spike that is going to happen in the next one hour, predictive DRS will detect it and will prepare the cluster to handle it.
Traditionally, DRS has always considered the compute resource (CPU and memory) utilizations of hosts and VMs for balancing load across hosts and placing VMs during power-on. This generally works well because in many cases, CPU and memory are the most important resources needed for good application performance.
However, since network availability is not considered in this approach, sometimes this results in placing or
migrating a VM to a host which is already network saturated. This might have some performance impact on the application if it happens to be network sensitive.
DRS is network-aware in vSphere 6.5, so it now considers the network utilization of host and network usage requirements of VMs during initial placement and load balancing. This makes DRS load balancing and initial placement of VMs more effective.
How It Works
During initial placement and load balancing, DRS first comes up with the list of best possible hosts to run a VM based on compute resources and then uses some heuristics to decide the final host based on VM and host network utilization’s. This makes sure the VM gets the network resources it needs along with the compute resources.
The goal of network-aware DRS in vSphere 6.5 is only to make sure the host has sufficient network resources available along with compute resources required by the VM. So, unlike regular DRS, which balances the CPU and memory load, network-aware DRS does not balance the network load in the cluster, which means it will not trigger a vMotion when there is network load imbalance.
Download a full DRS PERFORMANCE in VMware vSphere 6.5 Study Guide.
This animated video shows how VMware Storage DRS, a new feature of VMware vSphere 5.1 and vSOM, reduces the time and complexity of provisioning virtual machines by aggregating data stores into a single pool, called a datastore cluster, enabling rapid placement of virtual machines and virtual machine disks. VMware Storage DRS helps reduce costs and improve agility, balance storage space and I/O loads, and improve application performance. Learn about the new vCloud Suite: