VDI · Virtual Desktop
Virtual Desktop Infrastructure
What is VDI?
- Keeps desktops on central servers and sends only the screen
- No data stays on the device, which helps security
- Centralization means an outage stops the whole company
The first thing VDI changes is not security but how you manage PCs. IT teams used to repeat asset audits, OS patching, and antivirus deployment once per device. With VDI, you update one master image and redeploy, and hundreds of seats end up in the same state at once. Setting up a new hire's workstation drops from half a day to about ten minutes, and a lost laptop no longer turns into a data breach. IT's work shifts from repairing devices to managing server and storage capacity.
Even among VDI deployments, cost and user experience vary by type. Non-persistent VDI, which resets to a clean state at every login, is easy to manage as an image and saves storage, but programs users install do not survive. Persistent VDI, which assigns a fixed desktop to each person, is the opposite. For sizing, an office worker using documents, email, and ERP typically starts at 2 vCPUs and 6–8 GB of memory per user. Bandwidth is a few hundred kbps for document work, but it can reach several Mbps per user with video meetings or frequent screen changes. CAD and video editing need a separate pool with vGPU attached.
It also helps to draw the line against similar-looking concepts up front. RDS/SBC, where many people share one server OS through separate sessions, is cheaper, but one user's error can spill over to others. VDI gives each user an independent OS, which isolates better but consumes far more resources. DaaS is the form in which you rent this infrastructure for a monthly fee instead of buying it, and in Korea, logical network separation usually means putting either the business network or the internet network behind a virtual desktop.
The point where deployments most often go wrong is the assumption that 'we installed VDI, so security is done.' If clipboard, file transfer, and screen capture are left open, data walks right out, and photographing the screen with a phone cannot be blocked at all. Cost is also widely misunderstood. Once servers, storage, and OS and VDI licenses are added, the three-year total cost is often higher than buying new PCs. Capacity planning failures are even riskier. When a boot storm from everyone logging in at 9 a.m. on Monday makes logins take three minutes, business users quickly go back to their personal PCs, and if a single storage array goes down, every employee stops working at once.
Seungtae ParkHands-on AX instructor who teaches AI business development and marketing automation by building real systems
Ilyoung JungExpert in smart factories, equipment management, and Industry 4.0As virtualization technology for sharing server resources spread across enterprises in the mid-2000s, the approach of virtualizing the desktop environment itself and delivering it centrally became commercially available. In Korea, security regulations and network separation requirements greatly accelerated adoption.
At Company A, a precision-parts manufacturer with 1,200 employees and KRW 380 billion (≈ US$284,303,500) in annual revenue, an audit found signs that a departed design engineer had moved drawing files to a personal laptop. Drawings and cost estimates were scattered across the PCs of 400 office staff, and requests for remote work from on-site partner staff kept coming in. That is how a six-month VDI migration began. (Hypothetical example)
- At an executive meeting, the information security lead summed it up: 'If we can't trust the device, we just don't keep files on it.' That settled the direction: logically separated networks with VDI, instead of the plan to give everyone two PCs for physical separation.
- The IT infrastructure lead chose 260 of the 400 office seats, those running ERP and document work, for phase one, and set the 60 CAD seats of the design team aside in a separate vGPU pool. A design team manager had been blunt: 'If the mouse lags when I rotate a drawing, I can't use it for a single day.'
- For resources, we sized for 260 concurrent users at 2 vCPUs and 6GB per user, using 8 servers and all-flash storage. To avoid the morning login rush, we staggered each department's login time in 10-minute steps starting at 8:40.
- For export control, we opened the clipboard only from outside to inside and blocked USB. After the sales team complained, 'Can't we even attach files to email?', we added a separate file export channel that is processed within 2 hours of team lead approval.
- We ran a six-week pilot with 20 design team seats and received 17 complaints about screen lag. After tuning the protocol settings and the office network line, we expanded to the other departments in stages.
- Six months after the switch, endpoint incident tickets fell from 92 to 31 a month, setting up a new hire's workstation dropped from half a day to 20 minutes, and employees approved for remote work grew from 40 to 210. We made the storage redundant so connections stay up even if one side fails.