On Thu, Dec 11 2008 6:55 PM: dgrayson wrote: "Do PCs have a lower MTBF than thin-client? If they're a standard desktop, yes" - I don't know what thin clients you've been using, but I have found that exactly the opposite is true! I don't care what desktop you use. The user community will go through keyboards and mice on both, but the PC has a hard drive and, at least one fan. The hard-drive and motherboard are the two most common items replaced under warranty. I lost maybe one or two thin clients a year (usually because user buried it under paperwork & it overheated - or they spilled a Big Gulp on it)!
Frankly, I don't see any advantage at all to a one-to-one, thin client-to-blade configuration. The main reason why you see CAD and GIS or 3D applications in a Terminal Server environment is not because it makes sense, but because management figured-out that it's a convenient way to get a handful of seats on a single (read:expensive) license. The IT director looks like a hero for saving the company 20k and bringing them in to the virtual realm, while the rest of the team gets saddled with maintaining an elephant with a toothbrush and a cup-o-water.
To get back to the original question on objections. Yes, I have some objections to the Thin Client but you have to look at these scenario's in the proper context. Thin Client computing is an excellent solution for a workforce that lives in a single core business application with integrated reports and printing. As a Windows desktop replacement, it starts to encroach on the productivity of the end user. Mainly in form of application integration. As a common best practice (for security and stabilty of your farm), you'll probably have the clipboard, right-click, and <Alt>-Tab functions disabled. This severely handicaps the user from switching between applications, copy/paste, and a number of other activities taken for granted on the desktop PC and vital to any real productivity of your user community that lives in the MS-Office suite, for example.
So, while you can eat your beans with a knife, it is neither proper etiquette & perhaps just a little bit more dangerous than using the proper utensil.
Samuel A. RodriguezSr. Systems Administrator
The Cubix solution looks like a great workstation host, but it is limited to fibre network connections which are both expensive and distance limited (up to about 1mile). I am aware of multiple fortune 100 companies that are architecting new buildings to have network delivered desktops over a IP infrastructure (network switches with both copper and fibre links).
PC-over-IP (PCoIP) has been certified by multiple ISVs including the #1 CAD vendor Dessault Catia for the IBM HC10 and the Dell R5400. PCoIP was the first remoting workstation solution to be certified by Dessault System.
The feedback that I get from workstation end-users/IT is that they have looked at fibre and direct home run cabled solutions and there are significant challenges with the in-flexibility of the cabling and network architecture including moves/changes that can requiring patch panel changes etc. There is a strong preference for solutions that work on standard IP networks - which could include fibre, copper or wireless etc. The end customers I have spoken to included design engineers themselves as well as IT, CIO, CTO at many tier 1 car companies, aircraft manufacturing, architectural firms, healthcare (Xray, MRI etc), geospatial, oil & gas, scientific research, production studio, digital content creation etc.
In short, IT want a solution that can support long distances many 1,000's of miles - which requires an IP solution.
Some technical corrections on comments made regarding PCoIP Technology:
* PCoIP can run on any standard IP network and is not just over copper. Includes 3D graphics workstation operation over long distance networks - 5,000 miles is the longest test so far by an end customer that said they were "thrilled" with the performance.
* PCoIP supports full frame rate 3D graphics for any GPU with single/dual DVI links. DVI itself supports up to 1900x1200 resolution and 60 frames per second. While the majority interest is for single/dual DVI displays, we do get some requests for dual-link DVI which we will support in an upcoming product.
* PCoIP technology includes multiple capabilities to mitigate the effects of network latency - real-time network monitoring with dynamic quality adjustments to ensure the best user experience for a given network, broad dynamic bandwidth range from 100's of Kbps to 100's of Mbps depending on the available network resources, progressive image build that provides a fully loss-less/exact image of the host rendered image etc. More optimizations will be released in early 2009.
For full disclosure, I am the Director of Business Development at Teradici.
Sam - I think you need to review my statement which you quoted yesterday. I agreed with you that thin-clients have a better MTBF than standard PCs (lower MTBF is bad, higher is good), but I disagreed with you somewhat on the workstation-class PCs. Do drives and fans fail? Yes, but they also fail in servers feeding the thin-clients.
I do see why the overwhelming majority of 3D CAD, GIS, and other high-end commercial applications are run on standard PC workstation technology, and therefore, I do disagree with your second point from yesterday. In my (admitedly) limited experience with observing thin-client technologies operating in the context of professional 3D graphics environments, I have not been impressed by thin-client performance and other thin client-related issues, many of which are posted every day on this website's various forums.
Does this mean that I don't like thin-client or PC-Over-IP? Not at all - I use thin-client technology everyday in my own business. I do believe thin-clients will never be able to completely replace PC technology at the user's desktops due to the complexity of I/O-related processing issues and, more importantly, bandwidth limitations related to copper media. However, I see do agree that thin-client technology will continue to make advances into various commercial & consumer markets as the technology continues to mature.
As for Mr. Robinson's comments regarding Cubix technology, Teradici, and 3D graphics marketplace penetration, I'll leave any responses up to Cubix.
I'll retract or re-phrase then. (what you wrote seemed to indicate that you felt that "standard" desktops have a lower MTBF than Thin Clients).
Clearly you differentiate the "standard" desktop with a "workstation". And that makes sense. I agree. However, what I am saying that in todays terms, the standard dektop has the power of a Cray from just 15 years ago. What differentiates a standard desktop from a 3D workstation these days (with the advancements made in Processor, Memory Bus speeds, PCI-X or Express, and advent of SAS) is usually just the video adapter (Celeron notwithstanding) and amount of installed RAM. So, if you're a pure PC environment, you'll have hundreds of user with 20x more power than they'll ever need (degrading on a day-to-day basis), and the associated support costs I mentioned earlier.
The beauty is that if you do it right, you can provide a productive, feature-rich computing environment for the masses (x-percent of your user population) using TS/Citrix and Thin-Client appliances. Yes, there are tangible/intagible costs made clear by David. Like: specialized server skills, more server requirements, less peripheral support, & just acceptible multi-media support (beginning to be a problem). I mean, I have been in corporate environments where their desktop policies are as restrictive as my published desktops, so Thin Client may or may not affect user experience at all.
If you use the right tool for the job, you'll get the job done faster and more efficiently. e.g. Don't use pliers to torque head bolts & don't saddle power users, managers and the like with a Thin Client desktop.
I cannot comment intelligently on businesses whose core business is 3D graphics, Video, CAD, GIS, and other graphics intensive activities. My experience is in the a-typical business enterprise with maybe a department or two utilizing software like that. Geographical dispersion always presents challenges to any enterprise. Each architect has to determine the IT costs of how they conduct business, where it's most logical to save money and increase efficiencies. Not every technology delivers on it's promises, and like a stock portfolio - never put all your eggs in one basket.
Re: Thin Client objections?, and Mr. Robinson's opinions
Is there really any question of user preference here, “..broad dynamic bandwidth range from 100s of Kbps to 100s of Mbps…” versus 5Gbps, 10Gbps, or 20Gbps connection speeds? Gee whiz, I wonder which throughput choice most 3D CAD / MCAD, DCC, and NLE artists would like access to when working remotely from their workstation?
Cubix does manufacture an excellent workstation product as Stuart Robinson stated. However, Mr. Robinson is mistaken with his assertion that Cubix’s workstations require “fibre network connections”. Like most other PC solution builders, Cubix's blade workstation products feature standard, copper-based NICs, either built into the motherboards supplied by its vendors, or with add-on PCI or PCIe network adapters.
Cubix, along with companies such as Verizon and AT&T in their respective markets, chose fiber optics as the basis for its unique remote I/O technology due to the enormous bandwidth advantage that fiber has over copper media. End-users who do their homework also come to the conclusion that fiber is more cost effective than copper media in bulk deployments for supporting 3D graphics applications.
As for end-user preferences, the end-user prefers the solution(s) which will best meet the objectives of their organization. For some, this will be wireless technology. For many others, this will be IP-based copper technology. However, there are niche markets, such as professional 3D graphics, with applications which require much more bandwidth than copper-based media can provide for remote application deployments. The evidence is found in the concurrent growth, despite recent economic conditions, in both 3D graphics workstation demand and SAN technology demand in 3D graphics markets. In other words, the fiber HBAs which used to be deployed almost exclusively within IT data centers are now making their way out to the user workstations at graphics studios and engineering offices.
Cubix’s unique technology is pixel-for-pixel perfect since the original graphics data is actually processed at the user’s desktop, not back in the data center where the executable files reside. Cubix doesn't process video at the blade workstation, and then pump it out to the remote device. This is a very important point to keep in mind when evaluating the various technologies for comparison of strengths and weaknesses.
When streaming multiple 3D graphic frames at rates ranging from 24fps to hundreds of fps, copper’s bandwidth limitations, along with many visual & audio issues caused by the lossy compression algorythms inherent to PCoIP technology, becomes more pronounced and more difficult to overcome. The greater the resolution, bit-depth, and frame rate requirements, the greater the technical issues listed above.
Given a choice between technology which only supports up to hundreds of Mbps, and technology which features a selection of 5, 10, or 20Gbps bandwidth, which technology do you think the customer will choose? This technical advantage, along with many other features such as an open architecture and transparent USB2.0 /FW800 support, is a winning combination for Cubix when evaluated and compared head-to-head with thin-client solutions or PCoIP-based products such as the Dell R5400. Unlike some of its competitors, Cubix welcomes the opportunity to be evaluated and compared with other technologies.
For full disclosure, I am the Business Development Manager at Cubix Corporation.
Eric Carlson
Sam,
When comparing standard desktop PCs to PC workstations, graphics controllers are only one among several differentiators. Workstation builders will often integrate high-end chipsets and memory, HDDs, power supplies, and even more robust chassis fans or gimmicky cooling apparatus in order to provide customers with a more reliable and higher-performance solution than the standard desktop PC can provide. Are they always more reliable than a desktop PC or thin-client device? No, but the power and I/O flexibility features speak for themselves.
I agree with your statement regarding using the right tool(s) for the job. More often than not, this means requiring a mix of PC/thin-client or PC workstation/thin-client configurations at the workplace. Where I have an objection is when vendors claim a performance spec with their proprietary thin-client or PCoIP technology that simply is not achieveable, or sustainable, in real world applications such as 3D CAD/MCAD, the various digital content creation submarkets (DCC), Non-Linear Editing, and so on.
You are absolutely correct with your statement "never put all your eggs in one basket". In my opinion, this is what I hear Teradici and their partners urging both forum readers and clients to do when talking about their PCoIP technology - they can do everything a PC can do. I would not be surprised if it ran typical office and database applications as well as, if not superior to, many thin-client solutions. I would be very surprised if their technology withstood thorough evaluation scrutiny under real world application requirements for the 3D graphics markets that Cubix serves.
Eric Carlson, Cubix
To answer some of Eric Carlson comments on PCoIP.
You will require 5, 10, or 20 Gbps to remote a desktop (via PCIe) when you do not compress any of the data. When you compress data, specifically the imaging data, you can see improvements of 10-100x, depending on what you are sending across. This is the method used by the PCoIP technology. It takes graphics directly from the GPU, uncompromised, and presents them to the Portal device. This reduction in bandwidth allows one to go over a standard IP network via whatever medium that supports those bandwidths (copper, wireless, fiber etc)
If you require large bandwidths (multi Gbps), you limit the networking topologies available. The networking infrastructure becomes extremely expensive, and/or you must reduce yourself to a tethered solution (point-to-point connections without switches in between - which I believe the Cubix solution is - please correct me if I am wrong). The tethered solution may not seem like a big limitation, but for some companies, the ability to swap cubicles by simply re-configuring the settings of the portal (remotely or via connection-broker) is a very valuable feature.
I agree with Eric's comment: "When streaming multiple 3D graphic frames at rates ranging from 24fps to hundreds of fps, copper’s bandwidth limitations, along with many visual & audio issues caused by the lossy compression algorythms inherent to PCoIP technology, becomes more pronounced and more difficult to overcome. The greater the resolution, bit-depth, and frame rate requirements, the greater the technical issues listed above."
I don't know many users who require hundreds of fps to be visible on a display, so I cannot comment on that. Standard displays are within the 60 to 75 Hz (fps) range at 18 maybe 24 bits/pixel - which is easily handled by the hardware compression/decompression supported in the PCoIP host and portal devices. Most remoting solutions today are software based, and in that medium the audio/video lossy compression algorithms are pronounced. The PCoIP solution is a hardware solution that compresses the data dynamically in real-time, and then decompresses the data on the portal. When using the PCoIP device with a 100 Mbps connection (not railing this bandwidth), through a standard switch network will still provide you an uncompromised user experience at the portal, as verified by the CAD vendors themselves (ISV certifications).
I hope that Eric has a PCoIP solution in-house to test, evaluate, and compare against the Cubix solution. If he did I, don't believe he would be 'very suprised' at the capabilities that have been proven in several 'real world applications'.
Disclosure - I am a Principal Engineer at Teradici, and have been using the PCoIP portal with a blade PC for the past year as my main work computer. I have had zero down time with the solution, and have 2 displays 1600x1080 and 1440x900, both of which operate at 60 Hz, and at least one of them is in constant operation with 3D graphic programs being displayed. The other used for day-to-day operations.
I find it interesting that one of the most profound invitations to a discussion/debate of relevance to the decision-makers of so many IT shops out there was left to a couple of niche market vendors. Both products seem way-cool in their implementations, but IMO serve a niche market. I was hoping there would be some great counterpoint from those that manage a successful Thin Client environment of the garden variety. Pity.
IMO David Caddick presents the quintessential question for enterprises far and wide. Which one gets the job done and costs less? IT Support shops are continually being asked to do more with less. Particularily in the current economic times. The outlook doesn't appear to be moving in our direction either.
I think in many ways the debate would have probably taken on the "Linux vs Windows" flavor. As I peruse through many of the threads posted, it's becomes clear that there are a couple of camps in the Virtual Desktop/Application arena. Those that have got it right, and those that don't. Needless to say, the poor administrators that have attempting to make the square peg fit in the round hole, probably hate it (Thin Clients and associated server environment) & contemplate their choice of suicide on a daily basis. Thin Client computing takes a lot of commitment, planning, and testing. Those that decide over a weekend to implement one thinking it's as easy as installing Microsoft Office are in for a steep learning curve and most likely, a forfeiture of considerable personal time!
David makes some of the best arguments against the Thin Client. However, in the standard office environment There are a couple of bullets that aren't so true anymore.
"Always provides a better user experience" - In the case where the user "lives" in a single application, this isn't necesarily. true. In the case of published desktops, managing appropriate session-per-server ratio is important to maintaining good user experience.
"Better multimedia performance" - With the advent of 1Gbps to the desktop, the performance difference on a benchmark is phenominal, but to the human eye is usually negligable .
"More flexibility and Easier to Repurpose" - In a typical environment, I'm not so sure. Repurposing desktops in a typical scenario usually means either assigning machine to another emplyee, or a different mix of applications (desktop technician will have to physically move the PC, or perhaps do an attended install). In a proper Thin Client environment, the user can go to any device and get the exact same desktop, access to applications are as easy as adding the user to a security group (as long as the application has already been installed and published on the farm).
IMO - The larger the user base is and especially if they're on the same campus, the greater Thin-Client concept pays dividends. In terms of cost, you'll probably pay more initially, because the price of Thin Clients are on par with a std desktop PC. By the time those PC's are fully amortized, you'll be saving money.
My .02c (and then some) anyway,