Slashdot Mirror
Would You Move to Windows Thin Clients?
"Most users will be running basic MS Office apps, Groupwise for e-mail, and accessing some Oracle databases. A consultant hired for preliminary recommendations is saying that we should run Windows XP on the thin client boxes, not even the embedded version but the full one. Additionally, some of our users have more powerful applications like AutoCAD and ArcMap. We have already determined that those users will not be moving to the thin client machines.
Our department has spoken with a Citrix support/sales person who claims you can support up to 1000 clients on a single Citrix server. That seems so far from what I have generally read that I have a hard time buying it. Can anyone corroborate that claim? Again, most users will be using Office, Groupwise, and accessing Oracle DBs.
Does anyone have any experience with a workplace making this sort of migration? I would love to find a way to make it work, but from the research I have done so far, it doesn't look like we are going to get any cost-savings (unless they miraculously decide to go with Linux)."
The savings would be better with Linux, but they may very well be worthwhile anyway. Determine how much IT time you're going to save against the cost of the setup.
I never vote for anyone. I always vote against.
-- W.C. Fields
Okay, no I'm not.
... or at least, don't change it much. Maybe re-think your use of networking and other such things, but still within the framework of your existing OS setup.
Why change?
What do you need to 'fix' in the existing system?
Continue to use per-desk PC's, albeit cheaper/better/easier to administer, and continue to utilize 'upgrade' licensing schemes - the cheapest you can possibly get from Microsoft, is, probably, very cheap - to maintain the OS upgrade/fix end of things, and you can probably do this all very cheap.
"Thin-client" is just another word for extremely small/light/cheap PC-in-a-box. So maybe the simpler option to 'considering remote client solutions' is simply, put it all into better hardware (new CPU/more RAM/*standard* video for all systems/etc.) and don't change your existing software standards
If you maintain your existing stance with regard to how well your business problems are solved by your software systems, and give it a 'boost' with periodical hardware upgrades for key areas/servers every now and then, there is little reason to drastically change everything at all.
When the PC is akin to the size/heft of a block of printer paper, 'administration' becomes relatively trivial - particularly in an organization of 2,000 people or so... if you've got your software worked out.
Just get smaller, lighter, cheaper PC's, and refresh them every now and then. Convert as much as possible to cheap laptops and monitors, even.
They are out there.
; -- the corruption of government starts with its secrets. a truly free people keep no secrets. --
I deploy and support this type of environment for a living (until I can earn my living with Open Source). While the number of users per server depends largely on what applications that you are running, a good (conservative) average number is about 50 users per dual-processor server. I tend to deploy dual-processor machines as their are diminishing returns on quad-processor servers (For example going from dual- to quad- processor increases your user count per server from 50 to only about 75 or so). These rules of thumb are on your average Pentium III server with about 1.5 to 2 GB of RAM. The Office software and Groupwise will conform well to this rule of thumb. Not sure on the Oracle apps, but if they are well-behaved 32 bit applications (read no DOS, 16Bit) they will run fine. Obviously you will want to pilot this environment to bench mark your specific results. Servers with Pentium IV Xeons will probably scale much better.
You will definitely want Citrix here for the advanced management and capabilities over Terminal Services alone (application publishing, advanced load balancing, managment console, etc).
If you take the benchmark numbers I mentioned earlier and add 20% or so for redundancy, you are looking at a farm of about 24 servers vice 100. Using the management capabilities of Citrix and server cloning techniques, administration of this farm will be be pretty easy. A single, experienced Citrix administrator can handle most of the level 2 and 3 support for this farm. With server cloning, adding additional identical servers for growth/redundancy down the road is easy.
You have correctly identified users of AutoCad and ArcMap as poor candidates for this type of environment due to the heavy requirements and graphics of these applications.
I disagree with the consultant that full blown XP is the best solution for the client. He/she may be hedging their bet for any Windows based applications that would not run well under Terminal Services/Citrix. If this is not the case, there are several Linux-based thin clients that would work well and would have a lower cost.
No.
/* oops I accidentally made a comment, sorry */
I've been looking at this as well but on a smaller scale. First, Citrix doesn't support 10-25 users per server -- it's per processor. So a 4 way processor can conceivably handle 40-100 users and since you don't seem to be running hard core apps like CAD (you mentioned that these users would not be migrating), etc, let's say 20 users per processor or 80 users on a 4 way box.
That puts you at 12 Citrix servers.
Next, according to the Citrix folks I've worked with, Windows Server 2003 handles Citrix MUCH more efficiently than Windows 2000 resulting in -- according to them -- a doubling of the number of users possible on each server. Since I don't quite buy that, let's go with a factor of 1.5 times the users rather than doubling. But let's stick with 12 4-way Citrix server to account for redundancy which you will surely want for this solution. Heck -- let's go with 15 even. It's still a lot fewer than 100.
I agree that -- at the beginning, using old hardware would save money initially. But, consider the support angle for a second. Rather than new, identical thin clients, you'll still have whatever you currently support for desktops. If you just maintain the copy of Windows that's on there, when one breaks, it's a total reload. If you use the new thin clients, it's a matter of swapping out the unit and they're less expensive to buy initially as well as more reliable due to fewer moving parts. You should see support costs drop dramatically with this rollout.
The Citrix guy that mentioned 1000 users on a single box had to be talking about something much larger than 4 processors... personally, I would recommend a cluster of 4 ways servers for teh redundancy that it would provide in the event of a hardware failure.
Going with Linux won't necessarily save in the long run. Sure -- you'll save on the initial software acquisition. But consider the support, end-user retraining and other problems that could crop up. Even if you use Cross Over office or something like that, your users WILL require retraining and you will suffer a productivity issue initally.
Not going with a Linux solution won't automatically doom the project nor will it prevent savings from the implementation.
I'm all for Linux (I have it deployed where it makes sense), but am wary of making generalizations that it automatically saves money.
...at $WORK I have a Linux box that handles ~25 full screen X sessions all day long, and hovers around .10 load average. The box is an old dual PIII 500MHz with 2GB RAM, 36GB RAID 0+1 volume. All of the client machines are Windows XP desktops using VNC viewer locally to attach to Xvnc sessions on the Linux box (which is kicked off in xinetd on-demand).
The server itself was considered too slow to run Windows. It had been retired from the Windows domain, and the bean counters have written it off the books. Yet there it is, serving a vital business function to my users that is about to scale up to another ~30 users in the next couple of weeks.
Every bit of software involved in making this happen was free. The hardware was, effectively, free. And I'm already handling more users than the newfangled expensive Windows 2000 Terminal Server that is parked in the same rack.
This move was made to stave off the sudden onslaught of requests for a second machine (linux) at every desk. The corporate standard desktop has been and will be Windows for some time to come. But setting up a Linux box with lots of RAM, fast disk and Xvnc has already saved us over $45,000 that would have otherwise been spent on dedicated Linux machines at every desk.
I have four servers configured with a single PIII 1.4GHz CPU and 2GB of SDRAM in a Citrix farm supporting 1100 Mac and Unix users for MS Office apps, Lotus Notes and MS Internet Explorer. *My* Citrix sales engineer claimed that only 15 to 25 concurrent users can be supported per CPU, but I have found that 4x that is easily doable.
:)
Individual users have different ways of working --not everybody is going to slam the farm in the same way at the same time every time they need to use a published app. Even with all one hundred of my client access licenses checked out, CPU utilization and paging on individual boxes in the farm stayed well below my alarm thresholds.
Generic productivity apps for word processing, spreadsheets, presentations and email just don't demand much from a CPU. Given the apps you intend to support (Groupwise, MS Office, and some Oracle front ends) I can't see you needing even 10 Citrix servers. Your best bet is put up a test farm and then perfmon your CPU utilization and swapping. Baseline it and then start adding users. Citrix have a very liberal demo program and you should take advantage of it.
As far as corroborating the claim of 1000 clients on a single Citrix server, I supported over twice that many (2270 to be exact) UNIX users with a single Citrix server, a PII running at 233MHz with 768MB of RAM. I started with 15 concurrent access licenses and 512MB of RAM. I added 15 more access licenses and another stick of SDRAM after the early users started spreading the word to the rest of the users how useful it was to be able to take care of company documentation right on your HPUX box and not have to wait for one of the five bull-pen NT workstations to open up. I published the MS Office 97 suite, plus Lotus cc:Mail and Visio, and the server never bottlenecked at the CPU or in paging. I also got to surplus those 5 bull-pen boxes and save the company the annual support fee we were paying to our out-sourcer, half of which, btw, showed up in my xmas bonus that year.
If you are going to publish apps with low CPU utilization like Word and Excel, I think you can easily support your thousand users with a handful of Citrix boxes.
Let me reiterate one point: This is a user's perspective of Citrix from someone with a predominantly UNIX and networking technical background. I do *not* have a clue about the finer points of managing Citrix installations, nor do I wish to after the last few hateful months with it!
The big selling points of thin clients are supposed to be a lower TCO and better security and management. In short, for WinTerms at least, this is pure marketing bullshit. Sure, it's a couple of hundred bucks a seat cheaper for your hardware and your end users can't install a macro virus or whatever. Well, actually, that last point is wrong. It's perfectly possible to have a user trash your Citrix server if the code happens to get executed there because your AV vendor wasn't on the ball or a patch was broken, only now they can effectively take down twenty other users at the same time.
Another thing that they don't tell you is that software licensing is a fricking nightmare - 1,000 users and 500 seats still equates to 1,000 licenses in many cases. We use Microsoft apps a lot, and they are totally inconsistent in their licensing requirements for thin clients, so much so that we now have full time staff just looking into thin client software licensing issues. Some other vendors are better, others are worse. Others are MUCH worse. More $$$ on the TCO.
Let's look at that TCO a little, while we are here. A tier 1 Windows corporate PC (after bulk discount) is roughly 1000 for us, including all of your office apps (cheap because they are bundled). A Wyse term is setting us back around 700, including your hopefully per-seat software licenses (not as cheap because they are unbundled, but on a bulk purchase scheme). *But* for each 10-20 users, you need a server. We run at 15 and still have performance issues, and we are talking dual Xeon boxes with 2GB of ECC RAM here; not cheap. The hardware/software costs are, in fact, about the same per seat, if anything thin clients are more expensive.
So, that leaves the management aspect of TCO. OK, there's less patching to be done, right? Well, actually no, since all the updates on our traditional desktops and laptops are handled either by the AV application directly or via a systems management package. No savings there. Warranties? Nope. That evens out in the same way that the hardware costs do. First line support costs? No, users still have the same problems with Office and what not. Second and third line support costs then? Ah! Finally a difference; you get to cut back on all your school leaver PC techies at near minimum wage and hire some Citrix Consultants instead, sure you only need half as many, but they come at three times the salary.
Citrix itself comes across a horrible hack to anyone who has used UNIX thin clients over X11. Performance sucks if you try and do moving video; even VNC managed to do better. And by moving video, I mean a flash animation like you get in a web page, not DVD quality FMV. The screen update code is nothing short of appalling; quite often a webpage is unreadable because the *entire* screen is updating to display one lousy Flash advert, and I've even seen mouse rollovers on links cause this. Whatever happened to atomic updates? While I'm slating the Citrix code, lets take a look at some of the other issues I've had the misfortune to experience:
UNIX? They're not even circumcised! Savages!
Is thin client really a cost-saving approach to a large user environment?
In my experience, yes. Look at this:
At a customer site we have the following implementation, fully Linux: every location has one or more application servers where people log in with their thin clients. There is a master application server. At every location on application master is promoted to be a local master. Every night the local masters synchronize with the global master, and later the remaining application servers synchronize with their local masters.
This means that you have to install an application server only on the global master, the next day all other servers have that application as well. Zero point of administration when it comes to that. The configuration files are not synch'ed from the global master, but at every location the application servers synchronize their configuration (/etc directory with a few exceptions) with their local masters, which means for example that configuring a new printer just has to be done on the local master, the next day all servers at that location know about that new printer.
Users' home directories are mounted via NFS from another server.
And now for the Windows part in this picture: we use VMWare and their persistent disk images (I think they're called like that; they throw away all changes and remain constant). On these VMWares we run Windows 2000 servers to which people can connect via rdesktop (RDP). We are now able to administrate just the Windows image of the global master (by temporarily switching off the persistance option), and the next day all other Windows images are the same. That's also pretty resistant to viruses and worms: just reboot in case of infection :-)
This saves money, since the only points to administrate are the global master and sometimes the local masters (site specific configuration stuff like printers).