Slashdot Mirror
How To Create a Linux Network for Peanuts
securitas writes: LinuxWorld has the first installment of a series on how to go from being a Windows based shop to a Linux one." One of the article's points, one that I strongly agree with, is how overpowered the machines are that most people buy.
I thought that I might never have to hear or read that word again. The bad memories of downed networks because some user unplugged his machine or knocked off the connector or removed the terminator are still way too fresh.
Why can't we all just get along without it? Splurge the eleven dollars for a 10/100 NIC and put in CAT 5.
This article is way off-base on several points. If my employer suggested that I maintain a garage sale network as described, I'd find another job. Yes, X-windows terminals are a perfectly valid way to go, but put a halfway decent machine on the job. You and your users will be much happier.
"Computers are useless. They can only give you answers."
-- Pablo Picasso
Often times you simply cannot find cheap hardware to purchase, unless you want to build it yourself or go with refurbished units.
Build it yourself is a poor option because it is very hard to find the quantities of parts you need, especially since business environments value similarity in desktop platforms. So you end up with groups of five or ten machines with whatever was on sale that week at Fry's Electronics.
If you are like most Windows-based companies you turn to vendors like Dell/Compaq/IBM and then the problem is that the cheaper machine you can buy is still a 900MHz Celeron with 256MB of RAM and a 20GB hard drive (granted it's only $600 but still what if you just need it to run training applications through a web browser?). Plus since you are riding the tail end of the cost range, you again enter the problem of having a month go by and suddenly you have completly different hardware.
So it's a choice between
* one vendor to resolve problems
* one platform to support/rollout
* one price that's not so great
or
* many vendors fingerpointing each other
* need a different image for every 5th system
* a price hovering around the lowest possible
For home/small business users I think the second choice is a valid one, but for large business and corporations I just don't think they'll ever see the value in it.
- JoeShmoe
-- I wonder which will go down in history as the bigger failure: the War on Drugs or the War on Filesharing
I'm touching power.
Let's say for the desktop in an business, administration or education setting, there is no reason to ever have more than say...500 MHz.
Simply because Word,StarOffice,WordPerfect, IE,Nutscrape,Outlook,Excel or Palm Desktop do not need more than 500 MHz, heck...you'd be fine with 266 and enough RAM.
If I go from a 266 MHz to a 1.4 GHz...the only difference is going to be a little bit quicker opening time for the application, and if the application and OS are done right...you only do that once or twice a day.
For other...more specialized applications like graphics or database admin or development, a faster CPU is needed. But for the vast majority of desktops...a faster CPU doesn't give you that much more for the money.
Intel and AMD should have focused on cooler CPUs in the 400-700 MHz range that draw less power so better enclosures for limited space settings could have been developed. You know, little boxes like Apple's Cube or the iMac, but with Intel or AMD cpus for education and business.
One of the article's points, one that I strongly agree with, is how overpowered the machines are that most people buy.
Maybe if LinuxWorld got some decent powered machines, they wouldn't be Slashdotted already.
too many people wanting the latest and greatest. there are several people that i work with that use 300-400MHtz machines with no problem. how do they do it? they haven't fallen in the the MS/Corel/Intuit/'fill in the blank' propaganda trap of having the newest version.
i use quickbooks 1999!
it all comes down to understanding what you *need* to do.
there are people out there that need/deserve powerful machines and there are people that could be just fine with second or third tier equipment.
e.
This could be for a lot of reasons: mis-configuration, mis-design, software load on the system, bloat, whatever. There are users who are proad of the number of open windows they can have on a desktop, like this makes them a power user or something.
of course, there is the old "it's not a bug, it's a feature" factor as well"
Comparisons to known operating systems are obvious
- - -
Radio Free Nation
is a news site based on Slash Code
"If You have a Story, We have a Soap Box"
- - -
"It is a greater offense to steal men's labor, than their clothes"
As someone pointed out, build-it yourself generally sucks for any network with more than 20 computers or so. Finding antiquated parts in those quantities can be difficult. But for smaller networks, it's great. And as it happens, a smaller business or home or organization needing a small network is more likely to need to pinch pennies than a mega-conglomerate wih hundreds or thousands of machines, for whom such a setup would be too difficult.
Dyolf Knip
Gotta love the /. effect. I had a chance to mirror it quickly here.
Make sure you try the original link first, please - it seems to come and go
quis custodiet ipsos custodes - Juvenal
Usually more expensive, too. Check out IBM's thinclient lines. $500 and up, not even including the monitor. The upside of course is that you can get these pre-assembled and in any quantity, hard to do with 486SX machines. Overall, they're a good compromise between do-it-yourself dumb terminals and PC workstations.
Dyolf Knip
Let's cover the points on morale first:
Do you want a four year old computer on YOUR desk? Of course not. You don't care if the IT manager says that it meets your needs, you just want to get your work done as quickly and easily as possible. If I tried this implimentation in my shop, I'd expect to field complaints from dozens of users saying that "their e-mail and Netscape is taking too long to load". If they bitched loud and long enough, their boss will give them the 1.4 Ghz that they want, but not without giving everyone a bunch of headaches first.
Many of these people have faster computers at home, so they're used to having better desktop performance than what a Pentium 200 with 128MB of RAM can offer.
Now, the points on productivity:
Not only will these workers be very annoyed when a slow computer is put on their desk, but their work output will suffer as they wait an extra thirty minutes each day for their applications to load and to save their information. Most of these people are being payed $20+ an hour, so the cost savings from buying cheap equipment will be sucked up quickly.
Also, If the user is a current Windows user, they'll need to be re-trained for both Linux and it's office applications. It might over a week for the less-skilled workers to get the hang of it. While they are learning, don't expect them to be happy about this, either.
Older computers tend to break down more, as well, and without warrantees that support cost is coming out of the companies pocket.
In short, this article makes the critical mistake of putting your users FIRST when planning an IT solution. Keeping your employees/customers productive and happy is a LOT more expensive than most companies IT costs. If you try to pass off cheap PC's on your workers, you'll pay for it tenfold with creating tons of new problems.
Very likely. In this case, you can upgrade your server or add another to the network. I've seen load-balancing software for thinclient Windoze servers, I imagine it exists for *Nix systems as well. It's a hell of a lot easier and cheaper to add 512MB of RAM to one server than to run around upgrading every single user's PC.
Someone told me that in the DSL center for Bellsouth in Atlanta they keep track of what the lag through their system is. If it ever gets more than 40ms, they start adding routers. Same principle.
Dyolf Knip
So get a Pentium. There's not exactly a big price difference between them anymore. Personally, I'd spring for P2's or K6's since they'll be easier to find, about the same price, and these things'll have to be upgraded eventually.
What pisses me off is that 10 years down the road they're going to say "Man, that Athlon 1.4GHz isn't worth the silicon it's printed on. Maybe we can use it for an X-terminal..."
Dyolf Knip
In principle; I agree. For many office-related tasks.. these new Ghz P-4's are rediculous. Something a quarter of the speed would be adequate. HOWEVER...
Let's say you are setting up a new office. Where, exactly, are you going to buy those machines? You can't. If you buy old, used machines, your costs in maintenance go way up. You want a bunch of machines that are the same, it makes support much easier. A problem in one applies to all, and so does the solution.
So when you go out and buy 100 brand new mid-level dell workstations.. sure, you're buying something faster than you need... but you're buying them because they will WORK.
[snip]
IBM marketspeak
[/snip]
For big companies, NC's are the way to go.
I agree wholeheartedly.
Dyolf Knip
As a high-schooler whose summer job involved (among other things) a small-scale Linux deployment inside a ~200-person office, the strategies of doing so interest me greatly. However, I've always seen a few issues with remote execution and thin clients that I hope someone here with more experience can address.
/usr tree remotely? The senior admins won't let me do any of that if it will degrade the network.
There are three levels of remote management you can do: None, mounting certain directories remotely, and launching only an Xserver on the client. The main problems I've had with the second and third options are:
Does it take a substantial amount of bandwidth to mount (for example) the
Will the users notice the delays substantially on a 100Mb/s network? I understand that this may be ok for word processing, but some of our users (and the main reason why we have linux in our company now) are running airport simulation models that have a complex, graphics-heavy UI and generate reams and reams of data. Would putting apps like that across the network impede their performance substantially?
I can already ssh into our machines and make them run any program I've uploaded to a certain directory overnight. Are the maintenance savings really substantial enough to outweight the speed/bandwidth issues? Thanks.
Friends don't let friends misuse the subjunctive.
lots of AC's surpose schools out and lots of bord lawyers (-;
X get a card that has hardware acceleration is my advice that means one that has good support
(me I go for an S3 card every time as the old ones are well supported in XFree86 3.x)
realistaclly you want a window manger that is low on grapgics if you can get people to run TWM the better because that is rock solid and low bandwidth
(less XPM to shove across the pipe makes john a happy boy)
realistaclly this setup has been tried alot and works but really
how about storys about NIS and adding crypto into it
how about mergeing win2k and unix logins
lots of things I would like to know rather than beating the old TCO drum realistically who cares people go out and buy what they like in terms of cheapness whatever you want a bang for your buck then go down the tip and grab a machine put a free Word Processor on it and away you go
what really makes the differance is manageability why do you think everyone started going down the thin client route thats because its easy to manage and means less hassles and less hassles = cheaper
please stop trying to pull these stunts and try something out
regards
john jones
Here is why this doesn't work.
1)If you only have a few workstations for a lot of people, you are going to end up paying people to twiddle their thumbs while someone else types a memo. This negates any savings on having cheap/fewer workstations in the first couple of months.
2)Slower workstations are, well, slower. those small groups of 30 second waits add up. If you want an efficient office, you dont pay people to wait for their machines to load data/get email, etc. New hardware is dirt cheap right now. You can get a good 1ghz system for 700 dollars or so, so why not get one.
3)Older equipment breaks more and is harder to find parts for. Try to find 72 pin simms that are guaranteed to work for a decent price. didnt think so. How about bios issues with those old 2 dollar motherboards when you try to slap a newer hard drive on them. Digging up AT power supplies? Yes, they do still exist, but they are getting a little more difficult to find. and the pain of working with older machines when they break is hellacious. swapping the power supply in an ATX machine takes about 2 minutes. In an AT machine, it takes about half an hour since I have to pull the entire machine apart. And yes, I do do this regularly. I changed 2 AT power supplies last week. (I work at a Uni, not everyone has new machines).
4) Old networking sucks. One of the major points of having a network is having the ability to share files. This means you want switched 100mb everywhere. Again, it is cheap enough, why do you pay people to wait. Our main fileserver is on gigabit fiber, and we use it constantly. Copper gigabit network cards are coming way down in price right now, the switches are coming down soon, so you might as well be prepared to go gigabit when you need to.
5) No office is in a vacuum. Abiword and StarOffice may be great, but none of them read all Office file formats perfectly yet. You still need to use microsoft products to communicate with other offices, for better or worse. Not a troll, just the truth.
6) Outlook. omygod Outlook is neat. I never saw the utility of outlook and exchange until I worked in an office that used it efficiently. It is at the point where it is indespensible. The ability to share calendars, email, move files around, schedule meetings, etc is wonderful. Yes, this does mean you have to run NT and exchange on a sever, bt we have made this concession. With the exception of our exchange server and our pdc, we are all FreeBSD.
In an office of 20 people, a 1000-1200 bucks per every 2 years (our average upgrade cycle) for each person isn't a huge cost compared to the salary, electricity, water bills, etc. Why not spend that kind of cash to make sure that work can actually get done and you dont have a sysadmin running around saying "it almost works!" or here's a workaround.
I'm a unix admin, I hate administering NT, but I have no doubt as to its current utility in most work environments. The benefits it provides outweighs the costs of maintaining it, at least until the unix variants get up to speed on the capabilities.
Here's the specs:
- VIA Cyrix MII PR266 processor
- 64MB RAM (single DIMM and upgradable)
- 4MB Flash Memory Disk
- 24X CD-ROM Drive
- 56K Modem
- 10/100 bT Ethernet
- 2 USB Ports
- Keyboard & Mouse
- Speakers
- Linux and X boot CD with Realplayer G2, Netscape, Citrix and an IRC client.
BTW, Dyolf is right, except for the thinknic, thin clients [that I've seen] are $500+. Too rich for my blood.
pherris
"And a voice was screaming: 'Holy Jesus! What are these goddamn animals?'" - HST
Quibble: a 486 is probably too slow to run StarOffice. That thing is a beast.
I used to buy the very top of the line hardware and could never get enough power. A 386/33 was non-negotiable -- the 386/25 was just too weak. But now bottom of the line is more than enough.
More serious point: WHY WHY WHY are fonts so fscking hard on Linux? I've installed RH 5.2, 6.0 and just recently 7.1, and setting up fonts was different on each one, and always a black art.
StarOffice's cooperation with font servers actually seemed to take a step backwards at one point, and I simply stopped using it. Why don't modern Linux distributions just include the damned font server, at least in the "desktop" configurations? I understand they can't include the fonts themselves, but at least including the font server would be a great start. That is THE single biggest barricade to Linux on the desktop, given the existence of suites like StarOffice.
I don't have anything against network computing but I wouldnt buy 20 junkers for business use.
Lets look at your typical 486 beater you can pick up at a garage sale:
1. Nearing the end of its life cycle - that means better buy some power supplies that fit that 486 chassis.
2. You might not need much drive space but that 250 meg drive will be as sloooooooow. This may not be an issue depending on how much local drive use you expect.
3. Video cards. Your users are going to want to run at 800x600 or higher and those cheesy cards you find on a 10 year old machines won't cut it. Better buy some cheapo modern cards.
4. NIC, no biggie if you don't mind running at 10 mbit or using thinnet.
5. No USB, may or may not be a problem.
6. Floppy drives need cleaning/replacement if you want a dependable read/write. Floppies suck on new machines with new media let alone 10 year old boxes.
7. Keyboard and mice may not be to the liking of your users. I'm using a keyboard from a 486 right now on my Duron box and love it. Clean/replace mice is required in most cases.
That being said, in a corporate environment just buy the cheapest celeron or whatever to get some new equipment. For non-profits, hobbyists, communes, post-apocalyptic societies etc its a good idea but go for a Pentium level machine with some decent video.
The reason is maintanence. These machines don't require any software maintanence, because they don't have copies of any important software on them. All they've got is an X server, which doesn't change, and an OS, which doesn't change.
Applications are on a central server, with a SINGLE COPY, which is also easy to administer.
If you're having problems with bad hardware, then go ahead & buy out a large set of bottom-end dells for $600 a piece, just for the support if it makes you feel better. But then explain to your boss how you just spent $500 per unit for support for COMMODITY HARDWARE.
As for users, they have a job and it isn't to play around with their computers. The machine is a tool, and if it does what they need (word processing, spreadsheets, etc), then that's all there is to it. Besides, now you have centralized backups for all users. Imagine how much they'll appreciate that. AND it doesn't crash on them.
And on the ease of use side, how much user training does it take for the user to hit the 'k' button instead of the 'start' button? Hell one of the biggest criticisms of KDE is that it's too windows-like. Structure the menu any way you like (hell change the 'k' icon to say 'start' if you like -- open source lets you do that).
If it _REALLY_ bugs you to buy old hardware, then pick up some sun network computers. They're supported, easy to maintain (hardware wise), and have a really big name behind them. The machines will be so different that the users won't compare them to their PCs the same way.
Care about electronic freedom? Consider donating to the EFF!
Too much ANTI-MS BS, if this article was an editorial, fine, that I couldn't criticize, but if it was targetted for System administrators or people about to deploy a network in a small company, it litteraly missed the target.
:) ) But unfortunately, probably didn't archieve it's own objective.
1. Who cares how much ressource MS apps sucks and costs, if we are reading that article, chances are we already KNOW all that crap and are looking for an alternative.
2. About no one uses 386/486 anymore, writting a paragraph on how the pentium III are useless horsepower to run all these apps and a 386 would do fine is pointless, unless you plan to deploy a network in a 3rd world country.
3. It gives you pointers, nothing good for someone comming from a windows env. You want a step by step guide, sounding easy a-la-windows install, to make it look simple and straightfoward. That's the big problem with some linux article, the authors knows their systems so well, that they can't put themselves in the shoes of someone that install linux and doesn't know how to access his floppy from the shell because he's used to a:.
This is *NOT* a rant, but a constructive criticism about an article that attracted a lot of people (server was half dead
--- Metamoderating abusive downgraders since my 300th post.
\i{For instance, I had a P233 that ran WinNT just fine.}
You're a wimp. I ran NT4 server on a P-120.
General Relativity: Space-time tells matter where to go; Matter tells space-time what shape to be.
First, he isn't trying to stick it to MS. He is trying to save his company big bucks by using nearly free equipment and software.
That said, this really does hurt MS. They make lots of money off of cilent side software. While the server side software (W2K Server, BackOffice, etc) are very expensive, client outnumber servers by a huge factor. And W2K client + Office Pro is not exactly cheap per seat.
Also most trends move from the client to the enterprise server. As long as MS has a large foothold in the client market, there will be a push to migrate servers to it, out of a desire to unify platforms, simplify management, and use untrained lackeys as system administrators. These are mostly myths. Running an NT/W2K server is nearly as different from running office on W9X as Linux is, and knowing how to play minesweeper does not qualify someone to manage an enterprise server, but many times people making these decisions don't understand this.
Linux gaining credibility as an extremely low cost desktop solution that can be used by accountants, lawyers, secretaries, and other mostly non-technical users is a big win for Linux on the client and the server.
Yet he's using machines for terminals that are powerful enough to run StarOffice without any trouble. Why run the apps on the server?
There's an opportunity here. One of the remaining Linux players should build up a "Linux for business desktops" install, as a boxed product. Designed to install on low-end machines, it should install just the stuff needed by non-programmer business users, along with a suitable predetermined configuration with good security. Offer it as a boxed product, with one CD and one good manual, covering both the system and the office app, that's all you need to get work done. Offer a matching "Linux for business servers", with a compatible configuration. Sell through places like Costco and Smart and Final. Push the simplicity aspect - computers for business, without the bells and whistles.
While I don't disagree with your points, the main idea of this article is to use old PC hardware *as X terminals*, and having a half-decent modern machine act as the application server for these terminals.
I think this scheme could work, given two amendments:
-Use high quality, modern video cards.
-Use highest quality keyboard and mouse (you know, the latest and greatest logitech optical stuff)
-By the best monitors (at least 17", flat screen triniton sort of monitors)
-HIDE the ugly beige P100 from 1995 from the user.
I agree that I would be bummed out if a dusty old 486 or early pentium was sitting at my desk. I probably wouldn't work as hard. But, this way, they never see this ugly machine, and to top it off the components that the user is actually exposed to are top notch.
IT MOST CERTAINLY DOES NOT! I stand by both my link and my post.
quis custodiet ipsos custodes - Juvenal
I agree with you completely... I read that and went "my god". why would anyone in their right mind consider using that technology again? I still cringe thinking about trying to maintain that stuff.
:P)
For that matter, the disgustingly low cost of decent quality 10/100 pci nics (netgear comes to mind- I prefer intel or 3com, but cheap is cheap, right?) and the low cost of cat5 or at worst cat3 really makes thinnet an insane concept. For that matter, having all those collisions is not really my idea of fun- investing in a few decently priced switches would improve his network performance by quite a bit. (there's such a thing as LATENCY, besides pure bandwidth
It seems to me that the guy writing this article is some kind of nutjob just out of school or something, who sees a piece of crap PC and says "Hey! that'd make a great (DNS/DHCP/SMTP/whatever) server." and then he proceeds to build it, and go from there.
Now here's my take on it- if that guy worked for me, or I was hired to manage him, I'd fire his ass faster than you can say "GET OUT." people like that are dangerous, because they don't think about some of those important things... like stability, downtime costs, etc. I don't care of the bargain basement box was super cheap, I'd prefer to spend a few more hundred and be sure the damn thing will always run and be something I can get parts for if it breaks.
If I built my array of DHCP servers, or DNS servers, or something like that out of generic desktop 200-300mhz boxes (like he suggests) I would be gone. and I would deserve to get canned. to do that when you need to guarantee that things work is just blatantly retarded.
EOM
Also, X is chatty as we all know. If your network is already chatty, imagine running X over a 10 Megabit connection! My point is that desktops are over powered. They are supposed to be. If your ran all programs on the server, you are going to need a even more expensive server with scads of ram. With the desktops and some storage on the network via Network Attached Storage or a Storage Array Network, a few servers and a production system (Database, webserving...etc etc...) and you have a complete system that even if the infrastructure is down, is still useful. When the network goes, the users can still type up a letter, do a spreadsheet ...etc etc. They may have to save to harddisk and print and move it to the server later, but at least that time was not wasted. User older equipment DOES make sense though. If the user's ain't bitching about the computer being slow, then why replace it??
Gorkman
Nothing beats a S390 with a big batch of dumb terminals attached. Sure you only get text mode apps, but they were fast text mode apps and they worked. Currently my employer is getting rid of all those text mode apps in favor of web based interfaces which are noticably less responsive, noticably less featureful and oftentimes haven't been implemented yet. If you look at the implementation schedules, many of the teams on the project are already years behind -- many of them haven't even written a line of code yet. But we gotta get rid of the mainframes because the GUI's more user friendly and the web is the wave of the future...
I'm trying to teach myself to set people on fire with my mind... Is it hot in here?
Let them bitch. If they bitch too loudly fire them and hire new people who will actually work instead of bitching. People will bitch no matter what it's your job to make the company run not cave in to the demands of the clueless lusers. They have a job to do and so do you.
If you catch employees running winamp and wasting their time fiddling with their kids pictures then you ought to have a talk with them about working when they are work. If your employees are futzing around all day instead of working that's all the more reason to lock down their workstations and make sure only work apps are running.
War is necrophilia.
Why does everyone make it sound like you can't do this with an X terminal? Get a terminal with a sound card, you now have XMMS. They can change their background themes all they want to. To an end user, there wouldn't be that much difference. They don't know if their desktop comes from the machine on their desk or is beamed to them from Saturn.
These things would be perfect even for programming, really. Especially if you're into server side development and use Perl or Python.
If you need a sophisticated IDE then yeah, X terminals may not cut it. KDevelop should work on them. I'm not sure about Kylix.
Even things like the GIMP work fine over X over ethernet. I've done it.
But things like video editing and of course games would not be well suited to X terminals. But how many office users use (or should use) such things?
Having perused the article, many of his suggestions like using 10base2 and 386 computers are simply moronic.
He's penny wise and pound foolish.
The author would be a good name to put on a blacklist.. i.e. "Don't ever hire this guy to manage a network."
Pretty simple, actually. Open up your Sunday paper, pull out a pencil and start drawing. Soon Charlie Brown, Lucy, Snoopy and all the others will have PCs with our favorite Penguin on them.
--
"Outlook not so good." That magic 8-ball knows everything! I'll ask about Exchange Server next.
Just install it on server with Vmware or Win4Lin (server version) people who still need to run windows software can then log to these machines and run in.
The VUI will never take off in a business environment.
Can you imagine a room full of users sitting in their cubicals talking all at once to their computer all day long? Talk about annoying!
I can see it for control equipment, gaming, or as a personal novelty, but I just don't see it catching on. I don't see how it can. (and I'm not talking about technically)
load "linux",8,1
My understanding is that an X terminal basically has to do one job... display an image. The actual processing takes place on a central server.
If my assumptions are true, it wouldn't really matter what application you where using. Sure, if everyone where doing some high end stuff all at once you might put a pretty heavy load on the server...
Feel free to correct me if I'm wrong. I'm not as well versed in this as I could be, I know.
load "linux",8,1
Or in a closing time companion. :-)
But seriously folks, the catch in using older hardware is that the motherboard probably doesn't support large enough IDE hard drives for what you have in mind.
I see even classic Slashdot is now pretty much unusable on dial up anymore.
Just install these applicatiobs on a server with Linux+Vmware+WindowsXX and make your users connect to them !
Original centralized machines are (believe it or not) cheaper and yes, you can provide Sally the Secretary with a Pentium 133 as an X-Windows station. It's possible to do this.
But as several people have pointed out, just because you can doesn't mean you SHOULD. He makes a poor argument (other than cost) to return to centralized computing and several people have pointed out that, even if we ignore the advantages of distributed computing (there are several), company's are STILL willing to spend the revenue necessary for distributed computing.
In short, cost alone isn't enough.
A better argument would be to point out the advantages to centralized computing that are not cost related (mobile 'desktops', centralized administration (no more GHOSTING!), etc.) However, given management's previous experiences with centralized computing, this isn't likely to be persuasive arguement either.
An argument needs to be found that shows that Linux is cheaper and invokes the use of distributed computing. (The advantages of remote administration is a start - but there's a long way to go.)
Unfortuately, the sparcs can only run in 8 bit. Many apps look terrible. Have you seen Mozilla running in 8 bit? Its theme alone chomps all the colours not to mention the problems there have been with the I-beam becoming invisible. The mere mention of Java or Shockwave is enough to send the CPU guage completely into the yellow.
Also programs that seem fine on a local X seem to update so incredibly slowly they become practily unusable when run posted remotely to the sparcs (Abiword when tyring to wrap text to a new line was guility here but maybe it was just that early build).
Worse still getting StarOffice up and running was nothing short of a nightmare (it would just core dump whenever it was posted to the sparcs). In the end I managed to find an IGNORE_XSESSIONERRORS envvar which let users start it up (with a core dump left behind).
When it comes to defaults for new users there is trouble there too - I installed Ximian to let us run Galeon because Mozilla was too slow. Unfortunately Ximian's pretty installer defaults to using Nautilus which completely overloaded the server when one sesison was running let alone four or five (make it stop! I mean start)... I made a gmc setting but getting rid of Nautilus as the default desktop manager once it's installed itself isn't as simple as it could be. In fact, it simply isn't (yet) all that easy to set up sensible Gnome desktop defaults for all new users - simple things like turning off thumbnail updating is important because when several machines are doing it at the same time it drains percious cpu. Maybe KDE would be better but that seems to run even slower than Gnome.
I've had to eject countless disks remotely because users have put them into the eject buttonless sparcs not realising that they could only access the floppy drive on the server and have then wondered exactly how they get their floppy back.
The idea of using esound turned out to be a stumbling block due to broken esound on the sparcs (I've tired building cross compilers but they never seem to completely work).
I need serious convincing that a bunch of dumb terminals really are better solution. Todays apps need more bandwidth and CPU than ever and when it's being shared out over a compartively slow bandwidth everyone suffers. If everyone stuck to using xterms then it wouldn't be so bad...
Maybe things don't feel so slow on 10/100Mbit networks but people readlily point it out here (why does it take so long to login?) to the extent that I'm undecided whether using NT4 on a PII with 64Mb is actually any worse.
I'm hoping that was sarcasm. ;)
That would take a HUGE server to accomodate a handfull of people.
And how would licensing work for something like that? I'm not sure about vmware, but I know that Microsoft would find some objection to you installing one copy of Windows 98 and sharing it via Linux with 50 users.
load "linux",8,1
AutoCad: LinuxCAD at http://softwareforge.com/
Users that are used to AutoCad want to use AutoCad.
However, if they're using AutoCad, their workstation is a major tool to them and you could easily justify a hot one. The AutoCad guy would probably just use the Linux box for file storage.
load "linux",8,1
Take a workable alternative: buy a bunch of near-identical machines from an educational institution or business that's doing an upgrade rollout. Rip the hard drives out, upgrade the RAM, keep a one-line-per-machine config file that maps MAC address onto hostname, IP, kernel/filesystem components to mount for each machine (e.g. different kernel for each network card, different image mounted on /etc/X11 for each video card).
/etc/X11 images mounted in various combinations for thirty machines. Each machine (Digital Venturis FX-2, P2-133) cost $Oz100 (about USD$50) including screen plus $Oz80 each for a 256MB SIMM plus about 30 minutes per machine checking it out and recording the config (total of $Oz5400 ~~ USD2700 plus two full days for the machines themselves plus cables, cabling time, and two 16-port switches for 30 workstations). Swapping via the (CAT5 100MHz) LAN saves a machine from locking up when the user does something dim, but causes it to run slowly to let them know that they goofed).
In my case, I wind up with three different kernels and three different
If you build bitzer machines, be prepared for endless headaches making everything work together. If you use noname machines (and sometimes if you don't) be prepared to discover that not all of those crashes were Windows' fault. <rant>If Mr Trey ``It-Will-Work-Next-Version-For-Sure Oh-This-Blue-Screen-Must-Be-Why-Its-Called-Beta'' Gates had taken the time and trouble to sell reliable, predictable software instead of pushing pretty rubbish out the door in a flurry of false reassurances, the machines we use wouldn't be so crappy: blame for failures would land fair and square where it belongs instead of being masked behind said crappy software and the problems would actually get (ghasp) fixed!</rant>
Got time? Spend some of it coding or testing
Or even, specially not in the third world.
The Brazilian (ok, Brazil is not a third world country, but we are far from rich) popular computer project uses a AMD K6II 500 with 64 MB of RAM. Why? Probably because huge projects can not depend on the availability of out-of-line parts.
And the builders of this system agree with your third point also. They made an easy to install stripped down distribution based solely on a stripped down KDE (only Konqueror and KOffice, plus the supporting packges and apps). The workstation is diskless, with a 16 MB flashcard to boot from. All in all, it end up being a nice Internet/Office machine.
Absolutly not !
l /0 00673.html
u ar y/004941.html
/ 00 1385.html
Have you heard of the ltsp (www.ltsp.org) ? Linux Server Terminal Project, it's a very nice way to transform cheap and diskless PC into X-Terminal.
According to some people in the mailing lists :
http://www.ltsp.org/pipermail/discuss/2000-Apri
http://www.ltsp.org/pipermail/discuss/2001-Febr
http://www.ltsp.org/pipermail/discuss/2000-June
Vmware and Win4Lin run just fine in with ltsp. Although I concede that VMware server needs to be beefy. So, the trick is just to set somewhere a server with VMware or Win4Lin, and "occasional" wusers who can't do without their windows applications can log to that server ad execute them. See the Win4Lin benchmarks, they are interesting.
This of course, can be seen as migration process to Thin Client and a permannet solution.
I think the article is great. I think it makes perfect sense to admins who are concerned with saving costs.
Of course, I'd probably be fired if I suggested that any of my company's clients "share" workstations. Since most of them are in the business industry, they use their desktop or notebook to help portray their "executive" image. Somehow, I just can't see too many of our clients giving up their laptop for a dummy terminal.
And lets not forget that usually businesses bring visitors through their offices... lots of those visitors are potential customers or partners. NONE of those potential customers or partners is going to appreciate the technical marvel and efficiency of running p75s all around the office. It hurts your image to leave that crap laying around in plain sight, which is why most business managers who want to be viewed as "cutting technology leaders" are so interested in having nice shiny new PCs all around the office. It helps them make more money, and that's what they're all about.
You can argue this til the cows come home, but you might as well be arguing with a brick wall. I've seen the point of view that business execs use, and they certainly aren't successful because they're stupid. On the same level, when's the last time you ever saw a manager show up at work who wasn't dressed to impress? They want P4s, spotless datacenters, and the newest Windows on their desktops. Do they mind if their network admin puts Linux on the servers? nope, because they EXPECT servers to look foreign to them (usually). Enough rambling, I hope I've at least made some of you techno-freaks realize that technology is only a means to an end, nothing more.
Protector of Capitalist views,
Meorah
In any case, Linus must be well on his way.
--hongpong.com
This is assuming, of course, that the applications that the client want to run are supported by win4lin.
I may be wrong, but the last time I looked there where quite a few high end applications that would not work under win4lin. These are the same set of applications that have not been recreated for Linux.
If I am wrong, please tell me. I'd love to tell people that they could run QuickBooks 2001 and IE 6 under Win4Lin. It would make my life a whole lot easier.
load "linux",8,1
And here is an even less costly solution :
Windows NT + Windows Terminal Server on a server and http://rdesktop.sourceforge.net/ on the Linux client running on ltsp. rdp is a Microsoft protocol similar to ICA of Citrix Metaframe.
3. It gives you pointers, nothing good for someone comming from a windows env. You want a step by step guide, sounding easy a-la-windows install, to make it look simple and straightfoward. That's the big problem with some linux article, the authors knows their systems so well, that they can't put themselves in the shoes of someone that install linux and doesn't know how to access his floppy from the shell because he's used to a:.
What would be great is if someone were to put together a bootable CD iso that had the ability to search for dhcp servers and then Windows domains via SAMBA.
Something any MCSE could download and burn, then drop into any old PC with a nic and a CD drive.
Imagine you're a network admin with not a lot of time, you could hand one of these CDs to any new/visiting employee and just tell them to boot from it and use their normal password.
All you'd need is a Linux box sitting on the network somewhere running Webmin, for the Admin to add users to.
In fact, I'd love to burn such an ISO onto one of those 50MB business card CDs.
I can't be the first one to think of such a thing - I'm heading over now to ltsp.org to see if it's already available...
Cheers,
Jim in Tokyo
-- My Weblog.
I'm all for making Linux the enterprise standard, and I truly believe that there are a number of cases in which excessive computing power is used where not needed, but this article is a bit extreme. The auther leaves out a number of items which would be necessary to make this system work.
- Monitors. How are we supposed to look at pretty X widgets? Dot matrix printout?
- Network Equipment. A NIC card does not a network make. You are at least going to need some cable and hubs.
- Cost of installing the network. In most places where this solution is viable (small service businesses, order entry, churches), a network infrastructure is not in place. Files are passed on the floppy-net. Running cables on open floor is not an option, as it is an OSHA and fire safety hazard. So you either need to purchase and install raised floors, or resituate your offices.
- Scalability. The author never mentions the target number of users in this model. I can see this system comfortably supporting five users, possibly ten if all the employees need are simple text entry forms, but just try to run three instances of StarOffice and five of Netscape on the network, and watch your 300 MHz server grind to a halt.
- Progress. This system is great... if you believe your companies needs will NEVER change. There is absolutely no room for improvement here. What happens when each clerk must scan a barcode along with an entry? Do we ask the clerks to enter the barcode by hand?
- Customer/Employee satisfaction. No one likes to work on equipment that is known to be out dated and obsolete, even if it works well. That's why high school students bring graphing calculators to algebra courses. It would be very difficult to appeal to potential customers, no matter what business you are in, when you are using a system such as this. The same goes for employees.
The $30 system not only lacks many components, but even when flushed out would be hard pressed to find a viable business for implementation. The wiser systems administrator will allow for future growth, and be sure to catalogue ALL components of the system before making a proposal to management.
"The chatter from other offices is distracting"
I guess this says it all. Not only are the people doing the chattering wasting time but they are also distracting others from working.
The remedy should not be "The person who is being distracted can waste some more time downloading software, installing it, finding music ont he net, and then wasting bandwidth to stream to his ears". It should be "I hear a lot of chatter out here which means you guys don't have enough to do expect more work and tighter deadlines starting tommorow".
BTW it's one thing to hand a dilbert cartoon or to modify the background it's another to install some app which breaks your computer, consumes a ton of storage, or sucks bandwidth.
War is necrophilia.
They don't know if their desktop comes from the machine on their desk or is beamed to them from Saturn.
:-)
If it were being beamed from Saturn, they might suspect something since the latency would likely be a little on the high side.
Other than that, you're absolutely right. And given how much my SO complains about how slow her laptop is, and how little space there is on it, I'm thinking of sneaking behind her back and turning it into an pseudo-xterminal. It should also solve the problem of getting her access to all my mp3s.
"The urge to save humanity is almost always a false front for the urge to rule." --H.L. Mencken
I'd say the opposite is true in a properly set up thin client network. When someone's $100 computer breaks, you throw it out, and replace it with a new one. The data is stored remotely, so there's no need to transfer any files. Just replace with a new system and the user is up and running again. For the server side, you get much more reliability from 10 $100 computers than you do from one $1000 computer. There is a limit, of course, since you don't want to take up tons and tons of space. But again, if the system is properly set up, you throw out the nonworking computer and replace with a standby. If RAID is redundant array of inexpensive disks, I guess this would be RAIPC.
ok then your [sic] infringing on my copyright! Could you as [sic] me next time before STEALING my comments for your own?
I have tried it. I started StarOffice REMOTELY on my home machine from a university pool (on a HP/UX machine with 32MB RAM, that could just barely endure CDE). My home machine is in a dorm that is connected with 2MBit/s that is shared by 400 other people.
It was useable. Sure, you couldn't drag full screen images around but I could scroll down my text just as fast as I could locally. If you switched off the 'show content in moving windows' feature you wouldn't have known the difference.
I even started Win4Lin (www.netraverse.com) remotely once in fullscreen to shock our HP/UX admin. Worked as well (both ;-). Strangely, running Netscape on Win4Lin remotely over a shared 2 MBit/s line was more responsive than running a remote Netscape/Linux natively. I don't know what caused this.
But the fact is that you don't even need the full 10MBit to run a simple office application. Most of the time anyway.
Home Page
The article also said for the application server, it needs only be a 300Mhz computer with 80-256MB of RAM, which is pretty slim, too.
Engineering and the Ultimate
I didn't touch games because I was talking about the vast majority of computer purchases.
IMHO - 90 percent of computers purchased today have no need to be over 500 MHz.
SUN eventually decided that 100-base-T was the only way to effectively deploy that technology.
I used an X-Terminal for about 5 years to develop software. If I ran a local window manager, and did mostly x-terminals it ran smoothly -- graphics and complex GUI's are nowhere near as crisp and responsive. 4 trips up and down a network stack for every mouse click is much slower than anything Microsoft puts out.
Cat: The other white meat
I'm not perpetuating any myth; I'm saying that a 5mbit shared coax pipe across N users is going to have an increasing amount of contention for network space. The more you try to cram onto that large, SINGLE collision domain, the worse it will get. I remember those days and I'd personally rather avoid them again, considering the cost is pretty negligible.
:)
I'd also submit that having a 100-mbit duplex connection to a switch with that X server will definitely help X performance, especially as the number of users grows. but again, that's just my opinion. feel free to use archaic technologies and sell them as viable solutions all you want
EOM
The article is talking about a level of computing that I would call "Disposible". If you are dealing with terminals in the 300 mhz range, you can buy them in bulk on E-Bay or save them from being destroyed. These computers are out there and they are cheap... dirt cheap.
So if something goes bad, you don't call the vendor. You stick another one in it's place because you could afford to buy ten times the amount of hardware you needed.
About the only place I use thinnet anymore is a dirt cheap (charity job) link between buildings. Most traffic in internal to each building (print jobs etc.) so the slow traffic on the link is not a problem for them. Lightning protection is easier on underground coax. Use a pair of cheap 10 base hubs with BNC uplink ports and use a lighning arrestor on both ends. This is located in the telco room so it is not subject to the cleaning lady. Connect each to a port on your switch in each building. I don't recommend pulling CAT5 between buildings on seprate power grounds. It is too hard to protect. I do recommend fiber between buildings if you can afford it even though it is also 10 meg link speed.
The truth shall set you free!
What you're missing is that the author was comparing an X-terminal based Linux network with a conventional PC network. Running an office of PCs without a network is no longer possible, you need the network for printing and e-mail.
Thus, complaining about network gear, monitors and cabling is missing the point: both networks have similar needs there, so why discuss it? Maybe the author should have covered it in a sentence somewhere, since not everyone got it.
As for scalability, the X-terminal system is actually easier to scale, since you only have to upgrade the server, instead of an office full of workstations. And UNIX servers still scale better than Microsoft servers.
I don't know if X-terms can run barcode scanners, but I suspect they can. Just another input, like the keyboard and the mouse. You can certainly run the scanner off a serial port.
As for customer/employee satisfaction, how much is having a network that WORKS worth? One where the print queue doesn't randomly crap out on you (or the email server, or the proxy server...). Every network that I have ever been on EVER that ran off of Microsoft servers has been the victim of intermittent recurring failures in its components. They aren't down for long, you just get the reboot-monkeys to cycle the power on the machines, but if each system only fails once a month, that means something is probably going to go wrong every week or so. That's too often.
If you have to give the users whiz-bang to get them on board, buy funky monitors and run a nice theme on the desktops. Hide the PCs under the desks.
Jon Acheson
All opinions expressed herein are my own, and not those of my employers, who are appalled.
Are they X terminals or Sunrays? Just curious. The server-load for handling a sunray could be higher than that for an X terminal that can drive it's own desktop.
But that's beside the point; of course you can go that approach; but that doesn't solve problems if you require a windows environment.
Also... are you actually running 25 desktops off that sun enterprise 250?
25 instances of netscape? how much memory?
I'm genuinely curious.