Post by Narasimha Krishnakumar, Product Management, End-User Computing
Properly designing and sizing storage is key to successfully deploying virtual desktops. Transitioning from pilot to production is completely dependent on getting both the storage design and sizing right. If you get this wrong- you will negatively affect user experience, application responsiveness and your cost per desktop.
So how do you get it right?
Ultimately, you will need to ensure that your VDI environment is designed to support adequate capacity and performance. Translating a physical laptop/desktop environment to a desktop VM typically requires about 40GB of storage capacity and 15 to 200 IOPS . In addition, the IO pattern is made up of steady state events during which it is dominated by small block random writes and highly correlated peak events which can be dominated by either reads or writes. These shifts complicate storage sizing and design for VDI.
In the end, you have two ways to overcome these challenges:
- Leverage new product features
- Implement new storage architectures
- View Replica Tiering– View 4.5 enabled customers to take advantage of replica tiering. This feature was primarily driven by changes in the storage industry and the advent of Flash/SSDs as a Tier within storage arrays. With View Replica Tiering, customers can leverage their investments in Flash/SSD and get better performance in VDI deployments. At a high level, this feature allows customers to place replicas of parent VM’s on a separate datastore which is based on Flash/SSD drives. By placing the replica on faster tier storage, all read IO operations are directly served from the Flash SSD. This is highly beneficial in VDI environments, which tend to have highly correlated events such as boot and login storms that generate a lot of read IOs to the storage subsystem.
- Further, customers can set aside a high performance storage tier for addressing their peak read operations and focus on sizing for steady state IO’s to cut costs. Cost savings can be really significant in environments that use storage arrays with a combination of spinning disk drives and SSDs.
- An important point to note here is that View Replica Tiering is a feature that should be used in conjunction with View Composer Linked Clones.
- View Storage Accelerator (VSA) – For read performance acceleration VMware has introduced the View Storage Accelerator. With this capability, customers only need to set aside a small portion (100MB to 2GB) of ESX RAM as cache that can be used by View. The cache is a content based read cache and leverages the vSphere platform feature known as Content Based Read Cache (CBRC).
VSA is great for stateless (non-persistent) as well as stateful (persistent) desktops and is completely transparent to the guest virtual machine/desktop. What’s more-it does not require any special storage array technology and provides additional performance benefits when used in conjunction with other Storage Array technologies from partners like Nimble. More importantly, VSA can help customers drive down VDI storage costs.
These two features help customers with storage sizing. However, until recently these features could not be used together. With Horizon View 5.2, customers can now leverage VSA and View Replica Tiering side by side.
VSA and View Replica Tiering help address sizing challenges-but what about storage capacity? To help on this front, VMware Horizon View has introduced two new features.
1. View Composer API for Array Integration (VCAI)
In VDI implementations, customers have the choice of using a Full Clone (stateful desktop) or a Linked Clone (stateless desktops). Stateless desktops offer 2 primary benefits:
- Image management – Customers can centrally manage and update Windows VDI images–e.g. IT can update a Windows image to include the latest service pack and upon recompose, all VDI VMs are pushed the same update in minutes.
- Storage capacity savings – With Linked Clones, a single master image is shared by all the VM’s in a pool which eliminates the need for storing the same Windows image for each user.
Linked Clones allow customers to break down the entire VM image into different components such as the parent image, OS disk, user disk etc. In addition, many customers have successfully used Linked Clones with Horizon View Persona management, and stored user settings and persona on a separate CIFS share.
When customers create a pool of stateless desktops based on Linked Clones, the Clone (redo log-based file) is created by vSphere. However, storage solutions by various VMware technology partners offer the ability to create and manage efficient clones. In VMware Horizon View 5.1, we introduced vStorage Clone Array Integration (VCAI) to give customers the ability to offload the Cloning operations to a NAS storage array. Customers get the best of both worlds with this feature. They get the benefits of image management built into Horizon View and storage space optimization directly from our partners. This feature is still a tech preview and VMware is working with our partners to make this a fully supported feature in the near future. For more information on VCAI-read my earlier blog here.
2. Space Efficient Virtual Disks
With Horizon View 5.2-VMware has just introduced a completely new virtual disk format known as Space Efficient Virtual Disks. This sparse format is similar to redo-log based disks-but has the added benefit of growing and shrinking dynamically. The new vmdk format is applicable to both block based storage (iSCSI, FC etc.,) and file based storage (NFS) and leverages customer investments in partner storage technologies. What’s more-customers do not need to do any additional work to take advantage of this feature-it is the default setting for View Composer Linked Clone pools. With SE Sparse-customers can:
- Reclaim space – The sparse disk format with redo-log based Linked Clones helps optimize storage capacity. Over the life cycle of a VM, the redo-log-based Clone grows in size. This clone growth is primarily driven by end users downloading files, installing applications, deleting files and applications etc. When space is deleted on the VM, it never gets deleted from the hypervisor or the storage array. Space Efficient Virtual Disks provide IT with the ability to reclaim this space through a set of primitives, which frees up space all the way from the VM to the storage array.
- Maximize space efficiency – The block size of these disks is 4KB (as opposed to a 512 byte block size with redo-log based disks) and it aligns well with the NTFS block size and provides maximum space efficiency. In addition, it helps keep IOs aligned on most partner storage arrays.
In addition to these storage features directly available in VMware Horizon View, customers can also take advantage of new and enhanced storage architectures that help optimize performance and capacity while driving down the cost of VDI. Customers can now leverage highly optimized SSD based storage array architectures, hybrid storage architectures that support a combination of SSD and dense storage, as well as Virtual Appliances and local storage designs. The View storage optimization features complement and enhance these storage technologies and help IT drive down costs while delivering best in class end user experience. For examples and more information, check out VMware’s new fully validated vFast Track Architectures with storage partners such as Fusion io, Virident, Tintri, Atlantis Computing and Nimble Storage today.
VMware Reference Architect Tristan Todd discusses Storage Considerations for Desktop Virtualization in this short video: