VDI s Fatal Flaw V3 Solves the Latency Bottleneck A V3 Systems White Paper Table of Contents Executive Summary... 2 Section 1: Traditional VDI vs. V3 Systems VDI... 3 1a) Components of a Traditional VDI Deployment... 3 1b) Components of a V3 Systems Local Storage VDI Deployment... 3 Section 2: Scaling Traditional VDI vs. V3 VDI... 4 2a) Scaling a Traditional VDI Deployment... 4 2b) Scaling a V3 Systems Local Storage VDI Deployment... 5 Section 3: The Effects of Network and Access Latency on VDI Performance... 5 3a) Understanding Network Latency... 5 3b) Understanding Access Latency... 6 3c) Reducing Total Latency is Key to VDI Success... 6 Summary...7 Written by: Colin Kelly Jr. V3 Systems www.v3sys.com 1
Executive Summary Virtual Desktop Infrastructure (VDI), when done correctly, can outperform and provide a better end user experience than a physical desktop or laptop. However, understanding deployment architecture options, network latency, and storage latency are key to a successful VDI deployment. VDI solutions offer organizations many advantages compared to physical PCs including: Lower TCO over three to five years Decreased total energy consumption and heat output Easier for IT staff to deploy and manage large numbers of Windows desktops Increased security The challenge is, traditional VDI deployments suffer from one systemic problem network latency inherent within the architecture slows overall throughput causing a performance bottleneck. Although minimal during a proof-of-concept phase, this performance bottleneck exacerbates as users are added to a scaled traditional VDI deployment leading to: Slower desktop and application performance Unhappy, unproductive users Unforeseen, unpredictable, and costly network and SAN upgrades Failed VDI initiatives V3 Systems unique Desktop Cloud Computing Appliance architecture is designed to eliminate the bottleneck plaguing traditional VDI deployments. V3 Systems provides the fastest performing virtual Windows desktops in the industry while also offering an architecture designed to predictably and consistently scale without hidden costs or performance degradations. A V3 Systems deployment provides the missing link, allowing organizations to fully take advantage of VDI s benefits without sacrificing system performance and user productivity. www.v3sys.com 2
Section 1: Traditional VDI vs. V3 Systems VDI 1a) Components of a Traditional VDI Deployment reads/writes for all users' virtual desktops must travel over lower bandwidth, higher latency network Figure 1 illustrates a typical VDI deployment with separate compute and storage elements including the operating systems and applications for each user resident on SAN-based storage. Compute (host) Storage (SAN) Figure 1: Traditional VDI Architecture VDI architecture is designed to provide a method of deploying hundreds or thousands of desktops within one environment. VDI does this by taking the core components of the personal computer (compute and storage) and separating them into separate server (compute) and storage (SAN) components networked together. In a physical PC, the connection between its compute component (the CPU) and the PC s storage component (its hard drive) are on the same backplane. The CPU has direct bus-level access to the hard drive allowing for fast read/write times (i.e. throughput) when responding to requests from the OS. A traditional VDI deployment dictates the compute and storage components are, 1) physically separated from each other and, 2) the OS and applications for each user s virtual machine reside on the storage piece of the deployment (SAN). A typical VDI deployment must depend on a networked connection between the two parts. Even the fastest networked connection induces latency and slows down user perceived read/write times (throughput) between the CPUs and storage (remember the OS and applications for each user reside on the storage). 1b) Components of a V3 Systems Local Storage VDI Deployment In contrast to a traditional VDI deployment, V3 Systems deploys the storage component physically next to the compute component on the same backplane inside a server appliance (without a network or translation layer between the components). A V3 Systems solution addresses the traditional performance bottleneck in three key ways: 1. V3 virtual desktop host appliances contain local storage and compute components within each server appliance (no SAN required for VDI). The OS and apps for each user reside in local storage and are directly accessible to the CPU at the bus level via a PCIe or SATA protocols. This provides a level of throughput (read/write requests and actual read/writes) impossible for even the best SAN-based traditional VDI deployment. 2. All local storage within a V3 appliance consists of certified solid-state Flash offering the fastest possible reads/writes available in a desktop virtualization use case. 3. V3 optimizes the deployment of virtual desktops in several ways to fully take advantage of the direct communication of the compute and storage components contained within each V3 server appliance. The combination of these three methods results in a high-performance VDI solution that completely eliminates the traditional VDI performance bottleneck at scale while also eliminating the need for a significant SAN deployment or upgrade. www.v3sys.com 3
The combination of these three methods results in a high-performance VDI solution that completely eliminates the traditional VDI performance bottleneck at scale while also eliminating the need for a significant SAN deployment or upgrade. Figure 2 shows a V3 Systems VDI deployment with combined compute and storage elements. A SAN is not required since all users OS and applications reside in local storage. Most organizations deploy a lightweight SAN or NAS for shared documents or user data. Compute (host) CPUs have direct bus-level access providing local read/writes for all users' desktops Storage (host) Figure 2: V3 Local Storage Architecture Section 2: Scaling Traditional VDI vs. V3 VDI 2a) Scaling a Traditional VDI Deployment reads/writes for all users' virtual desktops must travel over lower bandwidth, higher latency network Since the OS and apps for each user s desktop in a traditional VDI deployment are stored remotely on the SAN, adding users to the system can be unpredictable, costly, and inefficient. Adding storage and bandwidth capacity to a SAN tends to happen in segmented thresholds requiring significant upgrades at unpredictable times. It is not uncommon during the scale process for an organization to suddenly realize it needs a $100,000 or $500,000 (or more expensive) SAN infrastructure upgrade to add users at certain stepping off points in order to not severely degrade performance for all users. Compute (hosts) Storage (SAN) Figure 3: Scaled Traditional VDI Figure 3 depicts a scaled out VDI infrastructure requiring additional hosts (additional compute components) and upgrades to both the bandwidth capacity (larger pipe compared to Figure 1) and storage capacity of the SAN (additional units of the storage component). www.v3sys.com 4
2b) Scaling a V3 Systems Local Storage VDI Deployment CPUs have direct bus-level access providing local read/writes for all users' desktops Compute (hosts) Storage (hosts) Figure 4: Scaled V3 Systems VDI With Local Storage Figure 4 shows a scaled V3 environment requiring only the addition of V3 appliances in order to accommodate additional users. Scaling V3 Systems VDI infrastructure is as easy as adding new appliances to an existing V3 deployment. With V3 Systems self-contained compute-plus-storage architecture, there is no need to upgrade a SAN infrastructure when adding hosts to the environment. Since the local compute and local storage components on each individual V3 Systems appliance communicate internally for system reads/writes, adding additional V3 hosts to a deployment does not present a negative impact on overall system read/write performance for users. V3 Systems offers multiple versions of its Desktop Cloud Computing appliance with configurations supporting a range of 40 to 250 desktops per appliance. This provides a method for organizations to easily add capacity for smaller blocks of new users at a predictable and fixed cost. Section 3: The Effects of Network and Access Latency on VDI Performance By definition, latency is a measure of the time delay experienced within a system or group of systems. The precise definition of latency is context-sensitive, meaning it depends on the particular system and time being measured Although latency can never be totally eliminated, it can be significantly reduced. To understand how latency impacts virtual desktop performance, it is important to review the two types of latency that most negatively affect VDI: network latency, and access latency. 3a) Understanding Network Latency Network latency is important because it affects the delivery of information to virtual desktops. Network latency is the time delay from when a data packet leaves the designated source until it arrives at the designated destination. The amount of network latency experienced is primarily dependent upon two factors: the throughput of the network connection, and the packet size. Network connection throughput is very important as it determines how much data can be simultaneously supported on the network. Packet size affects latency because as the packets get longer, they take more time to transmit. In addition, factors of distance and network types also can affect latency. Based on all these factors, networking can either be an enabler or it can quickly become the bottleneck. The average person is probably most familiar with the concept of bandwidth as the one advertised by manufacturers of network equipment. However, latency matters equally to the end user experience in the behavior of network connections that result in faster response times. Even the fastest performing SAN-based traditional VDI deployment requires the CPU to use a packet-switched network to access each virtual desktop s OS and applications. While using faster hard drives in a SAN will help reduce access latency, this cannot mitigate the effects of network latency inherent to the system. www.v3sys.com 5
In a V3 Systems deployment, each user s virtual desktop OS and applications reside in local storage thus eliminating any latency induced by a network. In virtual desktops, the most important factors affecting computer access time are in the host server, which is delivering or hosting the virtual desktops. These factors with respect to the host server include: the speed of the CPU and bus, amount of RAM, storage type, and proximity of high-performance storage to the CPU. In disk drives, disk access time is the interval between the time data is requested by the system and the time the data is provided by the drive. For spinning disk hard drives, disk access time is determined using a sum of: the spin-up time, seek time, rotational delay, and transfer time. Figure 5: SAN Traffic and Latency Illustration 3b) Understanding Access Latency A second type of latency that negatively impacts VDI is access latency. Within networks at each end of the journey, a data packet can be subjected to both storage and hard disk access delays. The time delay between when a request is sent to an electronic system and when the access is completed or the requested data is returned, is known as Access Latency. Access time is the actual amount of time it takes to complete access or return the requested data, and is the way that local desktops or laptop computers, hard disk drives, and solid-state disks are commonly measured. As a rule, when access time is increased, there is a corresponding decrease in performance, ultimately resulting in a poor end user experience. Understanding access latency is important because a reduction in access latency yields the opportunity to improve performance. Computer access time is dependent on several factors that can increase or decrease speed, including the CPU clock speed, RAM capacity, bus speed, interface speed, type of data storage (spinning disk drive or solid-state), proximity of high-performance data (OS and Temp files) to the CPU, and other factors. Because desktop virtualization does not require significant hardware on the client (endpoint) side, the hardware in the end user s endpoint is typically not a constraining factor in virtual desktop performance. Solid-state NAND Flash hard drives access data without the use of moving parts, so they do not have what we classically call spin-up time, seek time, and rotational delay; however, there is a delay from hardware signal relay (and buffering) that is sometimes described as seek time with a magnitude of ~0.1 milliseconds (ms). V3 Systems uses only solid-state NAND Flash resident locally in each host server appliance. The storage on each V3 Systems appliance is configured specifically to optimize performance for a desktop virtualization use case. V3 Systems configures its local storage without the use of a software translation layer allowing direct computeto-storage communication, ensuring the fastest possible throughput, and providing the lowest possible total latency. 3c) Reducing Total Latency is Key to V3 VDI Success Latency and performance are inversely related with regards to a VDI deployment. The understanding of how to reduce total latency from both the network and storage access perspective is one of the primary keys to making VDI successful. Latency can hinder a VDI deployment at many layers. A system can have the fastest storage available, but it will not matter if the storage is too far from the computing resources, either via a traditional network or via a distributed software translation layer for storage. Latency can kill your architecture; it can provide artificial or unessential roadblocks that simply cause a solution to fail. As you increase (or decrease) latency, you directly decrease (or increase) the performance of the computing resources, and ultimately the quality of www.v3sys.com 6
the end user experience. Many architects and virtualization specialists believe that the key to high-performance VDI is to use storage with high IOPs (input/output requests per second). A traditional SAN-based VDI deployment will offer approximately 20 IOPs per user using spinning disk hard drives to approximately 300 IOPs per user using solid-state drives. A V3 Systems local storagebased solution typically offers at least 5,000 IOPs per user far besting even the fastest SAN. Increased IOPs are helpful in many ways, but only after you have addressed the number one issue plaguing VDI deployments latency. Merely presenting more IOPs per user doesn t fully provide the missing link needed to eliminate typical SAN-based VDI bottlenecks. V3 Systems addresses both sides of the latency problem to provide the best possible end user desktop virtualization experience. Summary When architected and deployed correctly, VDI affords IT managers and their organizations many advantages. The allure of VDI incentivizes many IT departments to deploy VDI pilot programs. While it is relatively easy to architect any small proofof-concept VDI deployment to perform adequately, once the project advances from pilot to deployment stages a latency bottleneck almost always manifests itself leading to system-wide performance degradation. End users quickly become less and less productive as already-poor performance decreases with each new user. This one fatal flaw in the original VDI architecture often leads to the abandonment of once promising VDI initiates. A V3 Systems VDI deployment prevents the latency bottleneck from occurring in the first place. V3 s localstorage architecture was created from the ground up to avoid latency bottlenecks and ensure consistent, fast desktop performance for each and every user regardless of scale. www.v3sys.com 7