How to Optimize Virtual Machine Performance: Focus on the Disk Subsystem

Virtual machines (VMs) have become an integral part of modern IT infrastructure. However, it’s not enough to simply deploy a VM – it’s critical to ensure its optimal performance. This article focuses on optimizing the disk subsystem of virtual machines, as it often becomes a bottleneck affecting application speed. We will explore various methods and tools to maximize the efficient use of disk resources and improve the responsiveness of your VMs.

Table of Contents

• Choosing the Optimal Disk Type
• Host-Level Caching
• Tuning the Guest OS Disk System
• Using I/O Scheduler

Choosing the Optimal Disk Type

Choosing the disk type for a virtual machine has a significant impact on its performance. Traditional hard disk drives (HDDs) have lower access and data transfer speeds compared to solid-state drives (SSDs) and NVMe drives. In most cases, switching to SSD or NVMe disks will significantly improve VM performance, especially for tasks that heavily use disk I/O.

Disk Types and Their Applications:

• HDD (Hard Disk Drive): Suitable for storing large amounts of data where access speed is not critical. Examples: archives, backups, file servers with low load. Not recommended for databases, virtual machines performing intensive read/write operations.
• SSD (Solid State Drive): Significantly faster than HDDs, providing low latency and high I/O speed. Recommended for most virtual machines, databases, web servers where high performance is required.
• NVMe (Non-Volatile Memory express): The fastest type of drives, using the PCI Express interface, providing maximum bandwidth and minimal latency. Recommended for mission-critical applications that are demanding on the disk subsystem, such as high-load databases, real-time applications, virtualization with a large number of active VMs.

Virtual Disk Formats:

• Thick Provisioning: When creating a virtual disk, all the reserved space is immediately allocated on the physical storage. Provides predictable performance, as there is no need to dynamically allocate space during VM operation.
• Thin Provisioning: The virtual disk occupies only the space on the physical storage that is actually used. Allows you to save disk space, but can lead to reduced performance if dynamic disk expansion is required during VM operation. Recommended to use with caution, monitoring the fullness of the physical storage.

Optimizing Disk Choice Using VMware vSphere as an Example In VMware vSphere, when creating a virtual disk, you are given a choice between Thick Provisioning and Thin Provisioning, as well as the ability to choose a storage type (Datastore), which can be located on various types of physical disks (HDD, SSD, NVMe).

Example: Creating a Virtual Disk with Thick Provisioning in vSphere:

• In the vSphere Web Client, navigate to the virtual machine.
• Click «Edit Settings».
• In the «Virtual Hardware» section, select «Add New Device» -> «Hard Disk».
• Specify the disk size.
• In the «Virtual Disk Provisioning» section, select «Thick Provision Eager Zeroed» or «Thick Provision Lazy Zeroed». «Eager Zeroed» formats the entire disk immediately, which takes longer to create but provides higher performance in the future. «Lazy Zeroed» formats the disk as needed.
• Select Datastore located on SSD or NVMe for maximum performance.
• Save the changes.

Example: Monitoring Disk Performance in vSphere:

• In the vSphere Web Client, navigate to the virtual machine.
• Go to the «Monitor» -> «Performance» tab.
• Select «Disk» from the «Chart Options» drop-down list.
• You can track metrics such as «Disk Latency», «Disk Read Rate», «Disk Write Rate» to assess the performance of the disk subsystem and identify potential problems. High latency usually indicates disk overload or an insufficiently fast disk type.

Example: Changing the storage type (Datastore) for a virtual machine: If your virtual machine is using storage located on an HDD, you can move it to storage located on an SSD to improve performance. To do this, you can use vMotion (if possible) or Storage vMotion.

Important: When choosing a disk type and virtual disk format, consider the performance requirements of a particular virtual machine and available resources. Careful planning and monitoring will allow you to choose the optimal configuration and avoid performance problems in the future.

Host-Level Caching

Host-level caching is a mechanism that allows you to use the RAM or SSDs of the host server as a cache for the disk operations of virtual machines. This can significantly speed up access to frequently used data and reduce the latencies associated with reading and writing to physical disks. Effective caching can greatly improve VM performance, especially for data-intensive applications.

Benefits of Host-Level Caching:

• Reduced Latency: Caching data in RAM or SSD significantly reduces the time it takes to access data compared to traditional hard drives.
• Increased Throughput: Caching allows you to handle more read and write requests, increasing the overall throughput of the disk subsystem.
• Reduced Load on Disks: Caching reduces the number of read and write operations on physical disks, which extends their lifespan and reduces the likelihood of bottlenecks.

Host-Level Caching Technologies:

• VMware vSphere Flash Read Cache (vFRC): Uses the SSD of the host server to cache read operations of virtual machines.
• Microsoft Hyper-V Cache: Allows you to use the host server’s RAM to cache disk operations of virtual machines.
• Linux Cache: The Linux operating system uses RAM to cache disk operations.

Configuring Caching Using VMware vSphere Flash Read Cache (vFRC) as an Example vFRC allows you to use local SSD disks of the host server to cache read operations of virtual machines, which significantly improves the performance of applications that intensively use data reading.

Example: Enabling and Configuring vFRC for a Virtual Machine:

• In the vSphere Web Client, navigate to the virtual machine.
• Click «Edit Settings».
• In the «Virtual Hardware» section, select the virtual disk for which you want to enable vFRC.
• Expand the «Disk Cache Configuration» section.
• Check the «Virtual Flash Read Cache» box.
• Specify the cache size (in MB or GB). The cache size depends on the amount of RAM and SSD available on the host server, as well as the performance requirements of the virtual machine. It is recommended to allocate enough space for frequently used data.
• Save the changes.

Example: Monitoring vFRC Performance:

• In the vSphere Web Client, navigate to the virtual machine.
• Go to the «Monitor» -> «Performance» tab.
• Select «Disk» from the «Chart Options» drop-down list.
• Track metrics such as «vFlash Read Cache Hit Rate» and «vFlash Read Cache Read Latency» to assess the effectiveness of caching. A high cache hit rate and low read latency indicate that caching is working effectively.

Example: Clearing vFRC: In some cases, it may be necessary to clear the vFRC cache, for example, after changing the virtual machine configuration or to troubleshoot performance issues. To do this, you can restart the virtual machine or use the ESXi Shell command:

esxcli storage vflash cache reset -v <vm_name>

Important: When configuring host-level caching, you must consider the amount of available RAM and SSD on the host server, as well as the performance requirements of the virtual machines. Incorrect caching settings can lead to reduced performance. Regular monitoring and performance analysis will help you optimize caching parameters and achieve maximum efficiency. Also, make sure that the hypervisor you are using supports the caching function and that proper compatibility is configured.

Tuning the Guest OS Disk System

Tuning the disk system of the guest operating system (OS) is an important step in optimizing the performance of a virtual machine. Properly configuring the disk subsystem parameters within the guest OS can significantly improve data read and write speeds, as well as reduce latency. This process includes file system optimization, disk parameter configuration, and the use of special tools to improve performance.

Key Aspects of Tuning the Guest OS Disk System:

• Partition Alignment: Proper partition alignment ensures that read and write operations are performed efficiently, without the need to read data located at the boundaries of physical disk blocks. Incorrect alignment can lead to a significant decrease in performance, especially on SSD disks.
• File System Selection: Different file systems have different characteristics and are suitable for different types of tasks. For example, XFS and ext4 are popular file systems for Linux, and NTFS is for Windows. Choosing the right file system can significantly affect the performance of the disk subsystem.
• File System Parameter Configuration: File systems have many parameters that can be configured to optimize performance. For example, you can change the block size, enable or disable journaling, and configure caching parameters.
• Disk Defragmentation: (only for file systems that require defragmentation, such as NTFS) Disk defragmentation allows you to organize files on the disk so that they are located sequentially. This reduces file access time and improves overall disk subsystem performance.

Tuning the Disk System Using Linux (ext4) as an Example ext4 is a common file system in Linux, offering many parameters for optimizing performance.

Example: Checking and Correcting Partition Alignment:

• Use the fdisk -l or parted command to view information about disk partitions and make sure they are aligned to the boundaries of physical disk blocks. For example:

  fdisk -l /dev/sda

• If the partitions are not aligned, you can use parted to recreate them with the correct alignment. Warning: recreating partitions will result in data loss, so you must first make a backup.

Example: Mounting the ext4 File System with Optimization Options:

• When mounting the ext4 file system, you can use various options to optimize performance. For example, you can use the noatime option to disable updating the file access time, which will reduce the number of write operations to the disk.
• Edit the /etc/fstab file and add mount options. For example:

  /dev/sda1 /mnt/data ext4 defaults,noatime 0 0

• After editing the /etc/fstab file, run the mount -a command to apply the changes.

Example: Configuring File System Parameters Using tune2fs:

• The tune2fs utility allows you to configure various parameters of the ext4 file system. For example, you can change the file system check interval:

  tune2fs -i 0 /dev/sda1

  This command disables periodic file system checks, which can improve performance, but increases the risk of file system corruption in the event of a failure. It is recommended to use with caution.

Important: When tuning the disk system of the guest OS, you must consider the features of the file system and application being used. Incorrect parameter settings can lead to reduced performance or even file system corruption. Before making changes, it is recommended to back up your data and carefully study the documentation. Also, make sure that the latest versions of virtual device drivers are installed in the guest OS (for example, VMware Tools or Hyper-V Integration Services) to ensure optimal performance.

Using I/O Scheduler

The I/O scheduler is an operating system component that manages the order in which requests to read and write data to disk are processed. Choosing and configuring the I/O scheduler can significantly affect the performance of the virtual machine’s disk subsystem, especially under high load. Different I/O schedulers have different scheduling algorithms and are suitable for different types of tasks.

Main Types of I/O Schedulers:

• CFQ (Completely Fair Queuing): Strives to ensure a fair distribution of disk subsystem resources between processes. Suitable for most scenarios, especially when multiple tasks are running on a virtual machine at the same time.
• Noop (No Operation): The simplest scheduler, which simply forwards requests to the disk in the order they are received. Suitable for use with SSD disks, where the data search delay is minimal.
• Deadline: Tries to fulfill requests within a specified time (deadline). Suitable for applications that require low latency, such as databases and real-time applications.
• Kyber: An improved scheduler designed for modern solid-state drives.

Choosing an I/O Scheduler for Virtual Machines:

• For virtual machines working with SSD disks, it is recommended to use the Noop or Kyber scheduler.
• For virtual machines working with HDD disks and performing multiple tasks at the same time, it is recommended to use the CFQ scheduler.
• For virtual machines that require low latency, it is recommended to use the Deadline scheduler.

Configuring the I/O Scheduler Using Linux as an Example In Linux, you can change the I/O scheduler for each disk.

Example: Viewing the Current I/O Scheduler for a Disk:

• Run the command:

  cat /sys/block/sda/queue/scheduler

  This command will output a list of available schedulers, and the current scheduler will be indicated in square brackets. For example: noop deadline [cfq]

Example: Changing the I/O Scheduler for a Disk:

• Run the command:

  echo noop > /sys/block/sda/queue/scheduler

  This command will set the noop scheduler for the /dev/sda disk. Warning: this change will not be saved after rebooting.

Example: Changing the I/O Scheduler for a Disk Permanently (systemd):

• Create the file /etc/udev/rules.d/60-scheduler.rules with the following content:

  ACTION=="add|change", KERNEL=="sda", ATTR{queue/scheduler}="noop"

  This udev rule will set the noop scheduler for the /dev/sda disk when the system boots. Change sda to the name of your disk.
• Reboot the system or run the udevadm trigger command to apply the changes.

Important: When choosing and configuring the I/O scheduler, you must consider the features of the disk subsystem and application being used. An incorrect scheduler choice can lead to reduced performance. Before making changes, it is recommended to test different schedulers and choose the best option for your configuration. Also, refer to your operating system and hypervisor documentation for more information on configuring the I/O scheduler.