Issue link: https://epubs.iltanet.org/i/98946
VDI: Not Your Traditional Desktop Deployment and stability. If the system is too slow or does not meet expectations, adoption will be low and user feedback will be less than desirable. To address these concerns, a firm must select proper components and understand the varying loads generated by the VDI system during common work stages. These loads fluctuate greatly given the task, time of day and number of users accessing the system. The key technical design factors include disk (input/output operations per second or IOPS), processor (CPU) and memory (RAM). Also important is disk drive performance; most VDI installations hit the threshold of disk space long before processor headroom is maxed out. To ensure designs take into account normal workday tasks, system logins and massive boot storms, you can invest in a testing tool that measures the performance and scalability of centralized desktop environments, such as with virtual desktop infrastructure. Using one of these tools (e.g., Login VSI), you can benchmark, capacity plan, load/stress test and monitor change impact analyses on VDI installations. Shared storage was preferred in order to provide highavailability and motion between hosts, much like what we have come to expect in the virtual server world. In looking at VDI storage tiers, images (OS and applications) typically consume about 25 to 40GB, cache (page and temporary) range from 5 to 15GB and profiles (settings and data) seem to range from 2 to 5 GB, on average. When sizing for shared storage, consider this formula: Capacity = (I x S x (3V)) + (D x C), where I equals the number of image(s), S equals image size in GB, V equals versions on hand, D equals the number of virtual desktops, and C equals cache in GB. For example, a design with one 40GB image, storing three versions, with 300 virtual desktops and 6 GB of cache would require 2.16TB of shared storage. Disk IOPS are critical when sizing your VDI storage system. You must consider the three most common stages of work: A Formula for Success Below are IOPS calculation samples, based on a firm running 200 VDI desktop sessions: Using a typical legal desktop running Windows 7 Enterprise 64-bit, Office 2010, FileSite 8.5 and common legal desktop utilities, we tested for scenarios across the board and arrived at formulas which account for: • Number of users per server • RAM memory per host • CPU sockets and cores • Disk IOP performance In most of our designs, we have deployed either Dell EqualLogic or NetApp shared storage using 15k SAS or SSD hard drives. For servers, we selected either Dell R710/720 or HP Proliant DL360/380 systems. We have found the sweet spot for users per server to be approximately 50 to 75 (dictated by memory and CPU), given our typical N+1 design. • The working stage • Boot storms • Login • Working Stage: 10 IOPS x number of desktops at a minimum, so 200 desktops require 2,000 IOPS • Boot Storm: 7,500 IOPS per desktop x 200 desktops = 1.5m IOPS over a five-minute window = 2,500 IOPS • Login: 14,000 IOPS per desktop x 200 desktops = 2.8m IOPS over a five-minute window = 4,666 IOPS When calculating IOPS, consider the type of disk. You must understand the performance spread across the number of physical spindles available in an array. Below are some IOP speeds based on the most commonly available drives on the market. ILTA White Paper 25