Configuring a U170 Shared Computing Environment

Transcription

1 Configuring a U170 Shared Computing Environment NComputing Inc. March 09, 2010 Overview NComputing's desktop virtualization technology enables significantly lower computing costs by letting multiple users share a single PC. With the U170 product up to 10 users can share and simultaneously access a single PC. The concept of shared computing is not new, in fact if you go back to the early days of mainframe computing, multiple users shared a single system using basic access terminals. PCs forever changed the economics of computing allowing people to have the power of the mainframe on their desk. As technology has progressed, the PC has become so powerful that most users are now only using a fraction of the capability of the machine, resulting in wasted resources and wasted power. NComputing takes this excess power and shares it among multiple users, resulting in more efficient usage of resources and hence lower costs, easier set-up and management and a far superior environmental footprint. NComputing manages the shared PC resources with the vspace desktop virtualization software and distributes those resources to each access device. NComputing vspace software efficiently accesses the features and functions inherent in the host PC s operating system to distribute its capabilities among users with minimal overhead. Therefore every U170 station can see operating system responsiveness and application performance that is nearly the same or equal to that of the host PC. Setting up a multi user environment is straight forward but there are a few fundamental computing attributes that should be known and understood. The document will outline how to best configure a U-series environment by discussing the best configurations and topologies for 5 and 10 multi-user environments.

2 Determining Your U170 Environment There are a number of attributes in choosing the right configuration and topology for a U170 deployment; we will cover each of the following: - Topology - Host PC - Power: Powered hubs and cables - Host or system performance o CPU o Memory o USB bus - Application performance - Internet performance It is important to note that all of these attributes are interrelated and what is implemented on one attribute may affect performance on another, so understanding these computing fundamentals is important. It is important to read the Users Guide first; this document will not address install processes or procedures. The topology or layout of a U170 environment is very important as USB compliance, power and cabling/hub needs must be met. Please map out your topology in advance to make sure you understand what it will look like in advance, performance and power considerations (noted below) and what cables and hubs will be needed. We will provide a topology check list after we discuss the factors that can impact the topology and performance. Host PC The recommended host configuration is defined in the User Guide; please make sure to choose a host with enough CPU power (for video), memory for the number of users and multiple USB 2.0 buses (if needed). Remember to configure the host so that USB ports are not shut down as this will terminate any U170 active session. Power: Powered Hubs and Cables The USB 2.0 specification calls for 500 ma per port. Each USB device with a standard mouse and keyboard attached is expected to use about 400 ma of power (this includes a power allowance for the mouse and keyboard is up to 150 ma). Therefore there can be only one U170 per USB port. Use POWERED hubs when daisy chaining U170 s or when adding any additional USB devices downstream to a U170. The USB specification and all OS s allow only 5 hops on a port, so don t exceed this number.

3 Cable quality is IMPORTANT! Use only high-quality USB 2.0 compliant (not compatible) cables. We recommend a minimum size of 24 awg (power) and 28 awg (signal). Cheap cables can cause excessive voltage drops that can lead to system stability issues. The maximum length of a USB 2.0 compliant cable is 5 meters. A hub or repeater can be used to extend the cable to 10 or even 15 meters, but each hub (the U170 is a hub) or repeater counts as a hop and you can t exceed 5 hops on a port. CPU Performance: It is easy to check CPU utilization by using the Windows Task Manager in Windows. It gives clear and constant feedback on the percentage of processing power consumed making it easy to evaluate your environment to determine if the amount of processing power needed for a given set of applications is available. For the majority of productivity applications and many multi-media applications, most dual core processors provide more than enough performance to handle 6 or even 10 users at one time*. NComputing recommends you test your actual application set prior to wider deployments. NComputing s operating system testing with our vspace software has demonstrated excellent fair share processor utilization that enables all tasks to be serviced equally. This means that even if the CPU usage spikes to 100%, vspace and the operating system evenly distribute the available CPU capacity across all the stations. So in the cases where the CPU hits 100%, each individual station s performance slows on a sliding scale. For example, a user s perceived performance may move from 2.5 GHz to 2.0 GHz to 1.5 GHz and then spring back once the demanding task is completed. Additionally, those times when the actual processing requests exceed what was anticipated are usually very brief, and user performance normally returns to near the host PC s native speed very quickly. * Generally the most stressful CPU test for a U170 is video decoding at full screen. Playing a video stream is highly dependent upon the video file type; then somewhat on the video player (decoder) and screen resolution. We benchmark against D1 (or DVD) sized videos of up to a NTSC Wide-screen resolution of 720x480. Again most current P5 class Intel/AMD dual core processors can handle many streams, at once, if more are needed simply upgrade the processor. Memory: The second PC resource to check is memory utilization and it can easily be checked via the same performance tab within the Windows Task Manager. NComputing generally recommends provisioning the host PC with 2-4 GBytes of memory. Also check your paged memory usage to be sure it doesn t exceed the OS limits of 372 MB (XP) or 530 MB (Server 2003) (marked in red in the Task Manager below).

4 USB Bus: The U170 device connects to a PC through any USB 2.0 compliant connection method, directly through a USB cable included in the kit or a powered USB 2.0 hub. Modern computers normally provide 480Mbps or 960Mbps USB bandwidth depending on whether they have one or two USB buses. The table below describes the maximum bandwidth of the USB bus for modern PC motherboards. PCs that contain multiple EHCI controllers will have better video performance when running more U170 devices on a PC. Vendor Chip Set USB or EHCI controllers Maximum Bandwidth AMD SB MB/s Broadcom HT MB/s Intel ICH MB/s Intel ICH MB/s Intel ICH MB/s nvidia ION Series MB/s * Note: This is NOT a comphrensive list of chip set that have more than one USB bus.

5 Two buses are important if you are running > 4-5 users and you wish to maximize the video performance. With 4 or more simultaneous video s playing the USB bus bandwidth is pushed and it may impact video performance. If this happens the U170 will drop frames from the video. To find out how many USB buses your PC has, simply open Device Manager and view Devices by connection. The number of Standard Enhanced PCI to USB host Controller is the number of USB buses. In the example below, the PC has two USB buses. To take advantage of multiple controllers, connect any downstream hubs back to the PC directly rather than connecting all U170s downstream of one USB connection. Adding an additional USB controller card through a PCI slot to a PC is also an option. Depending on how many U170 devices attached to the PC and how many USB buses the PC has, there are typically no bottlenecks between the U170 stations and the host. Normal PC network performance rules apply to the host PC, and the most common limit is the connection bandwidth to your Internet service provider. However, this bandwidth is a shared resource for all users, whether using individual PCs or using a shared environment.

6 Application Performance: When setting up a multi-user environment, testing of your applications in a typical usage model should be conducted to determine that the configured PC s resources (CPU and memory) are sufficient to meet performance expectations. Any dual core system from 2008 and beyond should more than meet the vast majority of application performance needs for most educational or business users. The following data show how well 10 users can run on a standard PC available in the market today. If your testing indicates that a system s dual core processor does not meet your performance needs, then the option is to improve the resources in the host PC by going to a faster processor or using one with more processing cores. Internet Performance: There is really no difference between running 10 users on a single multi-user PC or running 10 stand alone PC s. Just make sure to connect the host to the LAN with the proper allowances for multiple users to your broadband services. Topology As mentioned topology of a U170 environment is very important and can be impacted by the items mention above; therefore we are providing the following topology check list: Host USB Compliance Extra USB Ports Power Cables Distance U170 Topology Check List Enough CPU power to run your desired number of video streams Two USB buses if >4-5 users running video simultaneously Make sure there are no more than 5 hops in the topology Maximum distance is 5 meters per hop You will need to add powered USB hubs to add extra USB ports downstream of the U170 as the U170 automatically assigns most downstream devices to the U170 session. There can be no more than one U170 device per USB port (and each port must support the specified 500mA of power). Use only powered USB hubs. Use quality cables Using quality cables don't exceed more than 5 hops on a single USB port

7 The following topologies give valid examples how to count a USB hop and how not to exceed them. We include two other valid topologies later in the document. This is a daisy chained topology of 5 hops. Note that the power hubs add the ability to hop and an extra USB port in the first two U170 devices. #3 #5 #1 #4 #2 This is a topology of only 3 hops. Note that the first power hub add the ability move up to 5 meters from the host and then split out four U170 devices (at up to another 5 meters). And the second series of powered hubs are optional but shown here add extra USB ports for flash drives, a printer or other USB devices. #1 #2 #3

8 U170 5-User Environment Example In our 5-user example configuration, we used an off the shelf Lenovo 3000 H Series 1 that sold for about $300. We connected four U170 devices, 5 LCD monitors, and Windows Server 2003 Standard Edition as the OS. Many PCs today have eight or more USB ports. There is no need for a USB hub if there are enough ports available. We connected the four U170 devices directly to four USB ports in the back as shown in the picture below. To demonstrate the performance, we started a 720x480 video in full screen mode on each U170 devices. Video decoding is a very processor, client and USB bus intensive task and therefore a good test to determine if all these attributes of the host system are working well. Once we launched the four video streams, we monitored CPU and memory usage from the host with Windows Task Manager. A 720 x 480 video clip typically represents a widescreen DVD quality image.

9 As shown from Windows Task Manager, the CPU usage is about 85% and memory usage is less than 600MB of memory while all devices are running full screen video at the same time. This shows that this low cost dual core CPU has no problem handing the four streams and that memory is NOT taxed in this scenario. Our topology for this five user environment looked like this: PC U170 Port 1 Port 2 U170 Port 1 Port 2 U170 Port 1 Port 2 U170 Port 1 Port 2

10 USB bandwidth utilization is not a feature of the Windows Task Manager so it is not an easy attribute to measure. The maximum specified bandwidth of USB 2.0 is 480 Mb/sec but in reality the practical useable bandwidth is about 400 Mb/sec. A typical U170 device runs at about 10Mb/sec at idle and a video stream running between the host PC with vspace and the U170 will take up to 130 Mb/sec of bandwidth to support up to 30 frames per second (fps). If the USB bandwidth does not exist to support 30 fps, then the U170/vSpace system will begin to drop frames in the video to make up for the lack bandwidth. Our testing has shown that about four full screen videos (depending upon the video format, size, decoder (player) and USB bus implementation) can play without a noticeable frames dropping. If your usage pattern is to use a lot of full screen videos simultaneously then we suggest you implement only up to 4 users per USB 2.0 bus to ensure better video quality. Therefore the best set up or topology for nine users is to obtain a host PC with two USB buses and to divide those buses among the users by having four users on one powered hub and five users on the second hub. Please refer to the previous table for chip sets with two USB buses or a USB 2.0 PCI card can be added to the host. A topology that looks like this: Each U170 kit contains one device that adds a new user to a shared PC. Up to nine U170 devices may connect to a PC and support ten users (including the host) simultaneously. The U170 combined with a modern PC provides the most affordable and powerful shared computing environment in the market today.

11 The following image shows an example U170 installation configured for 10 users: Of course, as with any PC, the performance for each user depends upon the host configuration and application needs of the users. Specifically, performance is dependent upon the individual host hardware, memory, CPU, applications being used, operating system and other factors. There is almost no discernable difference between virtual and real desktop environments as long as the two key resources of processor utilization and memory use do not exceed 100%. As long as these two key parameters are appropriately sized for the application, the performance for each U170 user should be comparable to that of the host PC. U User Environment In our 10-user example configuration, we used an off the shelf Gateway DX that sold for $789. We connected nine U170 devices, ten LCD monitors, and Windows Server 2003 Standard Edition as the OS. There are not enough USB ports on this PC for 9 U170 devices. We divided the nine devices into two groups. Five U170 devices are connected to a D-Link sevenport powered hub and four U170 devices are connected to a D-Link four-port powered hub. Since this PC has two USB buses, we connect the seven-port hub to the back of the PC and the five-port hub to the front of the PC to maximize the USB bandwidth. Each hub is connected with a 5-meter long USB cable.

12 Five U170 devices connected to a hub Four U170 devices connected to a hub

13 For the purpose of this exercise, NComputing ran 10 active users and a variety of different applications to determine performance. The hardware configuration and setup details are described in Appendix A. For office productivity, we ran 10 active applications comprising 28 separate instances across a total of 10 users. The applications are listed in Table 1. Table 1: Active Applications Application List Microsoft Paint Microsoft Word Microsoft Excel Microsoft PowerPoint Microsoft Internet Explorer Microsoft Outlook Express Roboworks WordPad VLC Media Player Mozilla Firefox With all 10 stations (or unique users) active, 10 productivity applications running, and a total of three multi-media video clips playing at native resolution (~720 x 480) in full screen mode and one video clip playing at normal mode (430x370), the environment provided excellent interactive performance with no discernable dropped video frames.

14 As you can see in the Task Manager Screen capture, processor utilization was well below 40% and the memory consumption was only a little over 1.5GB. This clearly shows that this quad core host system had plenty of capacity to handle this configuration. The total set of applications and video clips playing are captured in the matrix below: App1 App2 App3 App4 Multi Media Clip User 1 (host) Windows Task Manager User x 360 video User 3 Mozilla Firefox Paint Word PowerPoint User 4 Excel Word Internet Paint Explorer User 5 Internet Explorer Outlook Express Firefox Excel 320 x 240 video User 6 Roboworks WordPad Excel Internet Explorer User x 480 video User 8 Excel Word PowerPoint Internet Explorer User x 360 video User 10 Internet Explorer Roboworks Excel Outlook Express

15 Summary Please review the topology check list and diagrams to configure your U170 multi-user environment so it will work optimally. Once done the results from these 5 and 10 users scenario demonstrates that our U170 product can provide excellent performance for each U170 user on NComputing s virtual desktops. We demonstrated that a host PC can power up to 10-users simultaneously while executing productivity, office and multimedia applications with plenty of CPU and memory resources to spare. These environments demonstrate how NComputing vspace desktop virtualization technology can utilizes excess PC computing capacity and distribute it to many users. 1 Host Lenovo System Configuration: Intel Core2 Duo 2.53 MHz, Intel graphics Media Accelerator 3100, 2GB DDR 800 Memory, Microsoft Windows Server 2003 R2 SP2 Standard Edition and Office 2003 (system hardware cost was $300, excluding software and taxes). 2 Host Gateway System Configuration: Intel Core 2 Quad Processor Q8400 (2.66GHz, 4MB L2 Cache, 1333MHz FSB). 4GB DDR2 SDRAM Memory. Microsoft Windows Server 2003 R2 SP2 Standard Edition and Office 2003 (system hardware cost was $789, excluding software and taxes).

16 Appendix A Setup Details The tested hardware configuration consisted of: Host 1 PC: Lenovo 3000 H Series Host 2 PC: Gateway DX Nine U170 kit 10 LCD monitors 10 Keyboards and mice 1 D-Link 7-port USB hub 1 D-Link 5-port USB hub Two 5-meter USB repeaters Setting up this demonstration is very easy. You simply install NComputing vspace on the host PC; then connect each U170 device to a keyboard, monitor, mouse, and host PC (via USB 2.0 cable) and speakers and microphone to their ports.

17 Appendix B Benchmarking Comments NComputing does not offer terminal benchmarks but can provide some direction if you would like to try it yourself. Benchmarking a multi-user system is a challenging task. Traditional PC benchmarks often focus on processor intensive applications and attempt to max-out the system s CPU utilization to deliver a score or time taken to perform tasks such as 3-D graphics rendering or audio/video compression. Such benchmarks may have relevance to high-end users (such as gamers or video editors) but are not representative of typical office or school environment work loads. Other standard benchmarks that try to simulate office workloads generate scores that represent maximum utilization of CPU performance and I/O performance for peripherals like disk drives, but again, a higher score on these types of benchmarks may not necessarily translate into a significantly different user experience with normal day to day computing tasks. The typical PC spends most of its time waiting for the user to type, read screen text with basic graphics material, as opposed to rushing through tasks as quickly as the benchmark programs simulate. (To understand this you only need to watch one of the office mark type benchmarks executing; the screens flash by so quickly you cannot actually see what is going on. Nobody types that fast or switches screens that quickly in real life). Benchmarking a shared computing or desktop virtualization environment has never been easy. You cannot take a single-user CPU-intensive benchmark and run multiple simultaneous copies to get any meaningful multi-user results. (And these benchmarks will not run on terminals with multiple instances.) A better metric is to observe the end user experience when running a workload that is typical of what a user will be doing in normal day-to-day computing. The reason the CPU and PC vendors do not promote such benchmarks is that there would be little difference between today s PC and last year s model; because the system would be mostly waiting for the user to type the next keystroke or read a page just downloaded from the Internet. One methodology of evaluating the performance of a multi-user environment is based on measuring system utilization during a simulated set of office tasks that includes realistic delays between tasks, and proves that even a basic PC of today can easily support 11 users running common applications. The focus here is user experience centric, not CPU cycle centric, and we believe this is much more relevant to most actual user environments. Performance Metrics There are a number of other metrics that most users find relevant in real-world office and school classroom environments. i) System Boot time. How long does it take to get to a Windows logon prompt and access a usable desktop after powering up the system?

18 ii) iii) iv) Application and file opening time. How long does it take to open an office document? File save time. How long does it take to save your document? Internet page load time measuring this can be difficult since it is generally more a function of the network performance, not the PC itself v) Video playback performance. For example, does educational content play smoothly with synchronized sound? vi) System power consumption. How much power is consumed by the system on average? However, even in measuring these parameters, the goal is to deliver responsive performance, versus comparing absolute numbers. Will users really notice the difference between 31 second and 33 second boot times, or 1 second and 1.2 second file save times? While the host system runs through its simulated work load, we can observe the system resource utilization through the performance monitor. Ideally, CPU performance should average below 90%, memory utilization should generally be less than the physical amount of memory installed (no system page thrashing), and the network should be sized and configured properly for the number of users. If these parameters are maintained, an NComputing environment will give each user performance that is very close to that of the native host system. (Again, even if the CPU load goes over 100% the user performance is in most cases is still very good.) Thus, traditional PC benchmarks are generally not a good measure of real-world user experience and have been used by the PC industry to promote system capabilities not actually required for normal user computing. Furthermore, such benchmarks may not even work in a multi-user environment. However, if you run the simulated workload metrics outlined above you can demonstrate to yourself that NComputing vspace virtual desktops deliver a user experience that is close to that of the host PC.