VDI is dead! Or is it?
Most investors and analysts have given up on VDI. Earlier this year, I wrote an article VDI is an 8% solution. Stop wasting your time on it. However, many companies haven't written off VDI. They need to deliver Windows desktops to a certain segment of their population. An 8% for a company with 500,000 employees is still 40,000 desktops! These companies often have a small army of engineers (30-40) dedicated to keep the infrastructure operational and struggle with VDI performance and cost. VDI with legacy architecture is a problem at large-scale, but it's also a problem at 500 desktops.
In my blog series on VDI, I listed out the top 10 reasons why VDI failed to live up to its early expectations. Broadly speaking the problems with VDI can be classified into three categories:
High CapEx & OpEx of servers and storageVDI is synonymous with high server and storage costs. Data centers have historically been optimized for server workloads. Tens, maybe a few hundred virtual machines, each requiring tens of gigabytes of storage and doing mostly reads. A desktop workload breaks the data center. A medium-sized VDI deployment can have a few thousand virtual machines, each requiring tens of gigabytes of storage and doing mostly writes. A VDI workload needs both high performance Flash storage and lots of storage.
High OpEx of VDI softwareVDI 1.0 software was designed to work around the problems of the legacy data center. In order to reduce the amount of storage required, VDI solutions introduced golden images, app layering, dynamics desktops. In order to work around the performance problems of spinning disks, VDI solutions leveraged local disks on servers.
Poor User Experience for many classes of apps and on mobile devicesSince all applications in the desktop are remoted, the user experience of web applications, video apps, real-time apps, VOIP applications, and others suffer. Users cannot work offline. Users don't want a Windows desktop on their mobile devices.
VDI 1.0 was designed for a legacy data center - servers, SAN/NAS, spinning disks, and small amounts of expensive Flash storage. So while I was at Citrix, we (the entire ecosystem) spent a lot of time optimizing XenDesktop for both the performance and cost of the legacy data center. The problems with VDI and XenDesktop we tried to address were:
- Could we use inline de-duplication software like Atlantis to reduce the amount of storage required?
- Could we create a single golden image and assemble desktops dynamically? PVS and MCS were developed to solve these problems.
- Could we separate apps into layers and add them dynamically to desktops? We acquired Ringcube to solve this problem.
- Could we leverage local disks to reduce cost of centralized storage? XenServer implemented local disk caches to enable this functionality.
We couldn't fix VDI directly at that time to improve performance and cost. So we cobbled together or built a number of new technologies to handle the limitations both in terms of cost and performance of legacy data center architecture. These technologies added significant operational complexity to a XenDesktop deployment. We hear horror stories from companies today who have deployed these technologies. The result of the complexity introduced by the number of moving parts and pieces is brittle deployments requiring a small army of dedicated engineers to keep them up and running. Read more about the VDI performance and cost challenges we tried to solve.
Unfortunately, five years ago there was no better option. Five years ago, the data center architecture and technology maturity did not permit a simpler solution. However, a set of new technologies has emerged: (a) the cloud, (b) control plane architectures (Meraki, Palo Alto Networks, etc.) and (c) hyper-convergence solutions (Nutanix, Simplivity, Atlantis). These technologies dramatically change how companies run data centers as predictable web and cloud scale enter the data center. What could VDI look like if we leveraged these new technologies? Stay tuned to find out...
Learn more about the VDI performance and cost challenges we tried to solve in my eBook, 'A Brief History of VDI 1.0'
Long live VDI 2.0!