VMware vSphere: VM Memory Issues Can Hit Your VM Performance
Virtualization is being widely used in modern data centers and cloud environments due to flexibility, reliability, agility, and ease of management. Its hypervisor abstracts the underlying hardware and transparently allocates to the Virtual Machines (VMs) independent of their Operating Systems (OS). With help of a hypervisor, multiple virtual instances or VMs can be created and run on a single hardware.
VMware vSphere offers the world's best server virtualization capabilities that offer running multiple virtual instances on a single hardware or physical machine. Failure of hardware or poor performance of a physical machine affects the applications running in VMs on that machine, and with virtualization technology, multiple VMs running on that particular physical host can slow down the host and affect applications running on all of the VMs. Hence, performance monitoring of a virtual infrastructure is also very important to smoothly run and manage it.
The performance of any app that is running on a VM in a VMware vSphere infrastructure depends on many factors including:
Server virtualization abstracts the physical resources such as CPU, memory, disk, and network from the underlying hardware or hosts, and shares them between hosted VMs on that hardware, and if a few VMs consume a heavy amount of resources, then other VMs can't have access to those physical resources when they need it; this affects the performance of the application on other VMs.
Administrators can limit the available resources on installed VMs, and not properly configured limits can congest the performance of applications on those VMs.
VMware admins sometimes over-commit physical resources to VMs on the hosts, and not all VMs on these hosts may have the requirements of physical resources at the same time. Over-commitment of the physical resources ensures better utilization of resources, but it needs to monitor the actual utilization of resources on the hosts to identify where an ESXi host is resource-starved. If you haven't identified the actual utilization of resources, the performance of VMs may be affected.
To improve the performance of a VM, a virtualization engineer should track the usage of resources by a VM over time and right-size the resources of VMs accordingly.
VMware vSphere contains various VM memory access resource components including memory ballooning and memory swapping, and in this article, we'll discuss these memory access resources:
Memory ballooning in VMware vSphere environment is a memory reclamation mechanism used by the ESXi (Type-1 or bare metal) to allow the ESXi host to reclaim unused memory from VMs, where VMs that are experiencing a shortage in memory can use the reclaimed memory to run an application.
Normally, the hypervisor is responsible for assigning a portion of the memory of the ESXi host to each installed VM on it, and the guest OS that runs inside a VM is unaware of available memory to the ESXi host. Memory ballooning makes the installed guest OS in a VM aware of the shortage of the ESXi host's memory.
Whenever an ESXi host is in a situation of memory-contention and the VMware Tools with the ballooning driver are installed in the guest OS of the VM, they determine whether unused memory of any VM can be reclaimed. Then, the drivers inflate (balloon) the memory on an under-utilizing VM and prompts the ESXi host to reclaim the unused memory from that VM into the overall resource pool. The ESXi host then makes this additional physical memory available to other VMs on the ESXi host.
How Does Memory Ballooning Work?
Memory ballooning in VMware vSphere allows efficient utilization of physical memory of the ESXi host, however that comes at the cost of VM performance due to excessive memory ballooning on the ESXi host that enables the guest OS on the VM to read from the disk, and high disk I/O degrades the performance of a VM.
To prevent this excessive memory ballooning, VMware administrators should continuously track and monitor how much memory the ESXi host is reclaiming from the installed VMs and ensure not to grow too close to the set ballooning target. In the guest OS, monitoring of VMs offers little help in this regard, but monitoring the ballooning at host or hypervisor level proactively detects and controls memory excesses.
CNIL Metrics and Logs offers to monitor the memory usage, memory ballooning, and memory swapping of a VMs in VMware vSphere environment.
Sign up for CNIL Metrics and Logs to check your VM memory situation for free!
To find the above-mentioned important metrics, go to VMware Starter -> VMware Virtual Machine Dashboard -> Virtual Machine Memory Access Slowdown indicators.
Memory swapping occurs when the memory state of the physical host or ESXi host in VMware vSphere is under stress, or when reclamation mechanisms such as ballooning, compression, and page sharing are unable to keep pace with the rate of VMs memory allocation.
How Does Memory Swapping Work in Guest OS in vSphere?
Memory swapping occurs in VMware vSphere at two levels:
Memory Swapping at Guest OS Level: When a CPU accesses a virtual
memory at the guest OS level, it is swapped into physical memory and therefore virtual memory pages become available in physical memory and are served very quickly in the virtual environment. Memory pages which are rarely used are swapped to storage, and this creates a risk of high disk I/O resulting in slow overall application performance.
Memory Swapping at Hypervisor-level: At hypervisor-level, memory
pages on the VMs are swapped out to swap space, and each VM is linked to its own swap space. VMware vSphere manages the memory swapping from swap space when the guest OS accesses a memory page from the swap space. Swapping is always the last resort for a Hypervisor as data is moved from ultra-fast RAM to disk (SSD won't help much). Ballooning is always preferred to clear memory congestion. Therefore, active memory swapping at the Hypervisor level can also be caused by missing VMware tools in the Guest OS (that perform ballooning)..
Monitoring solutions that are available in the market and focus on VM performance can track the poor performance of a VM but aren't able to diagnose its root cause. An ideal monitoring solution for a VMware-based virtual environment will track memory swapping at the guest OS level and the hypervisor-level, while auto-correlating the metrics, and accurately pinpointing what is affecting the performance of a VM.
CNIL Metrics and Logs is one of the best VMware vSphere monitoring solutions that runs in minutes in your virtual environment and immediately starts monitoring the VM memory and other important metrics.
To improve VM performance in a virtual environment, VMware admins should track the version and status of installed VMware Tools on each VM, and the usage of CPUs and vCPUs can also be monitored so that both the physical host and VMs are right sized with the right CPU resources.
CNIL Metrics and Logs is a software that runs within minutes in your VMware-based virtual environment and immediately starts monitoring the VM memory access slowdown indicators and tracks these metrics over time.