Slashdot Mirror
Ask Slashdot: NT to Linux Migration Costs?
Alex asks:
"How would one go about calculating immediate costs and
tangible benefits of migrating an NT/Novell IntranetWare
LAN to Linux. I am not talking only about long-term
benefits as in more uptime, fewer crashes, robustness,
etc., but if I could use fewer and cheaper boxes for file,
internet and print servers. What could I say to a company
about the immediate benefits?" What about the costs
associated with retraining? What pitfalls do NT admins have
to worry about when moving to Linux?
I'm currently involved in an NT to Linux migration. I can't tell you what company this is for, but I will estimate for you the costs of this migration.
:-) Each machine in the cluster is a quad P-II Xeon with 1 GB RAM and 100 GB RAID hard drives. The system is kept together by two very hard working full-time staff. These two people are effectively on-call 24 hours a day. It has a team of six people in total managing technical and web design aspects.
Firstly - the service. The service is a popular European web-based application, mainly targetted at business users - shared email, calendaring and that sort of thing. It has a user base expected to increase to several million in the next year. It currently runs on a small cluster of NT servers fronted with a Linux/Apache proxy server (the Linux proxy server is another story - we use mod_rewrite to work around a number of bugs on the NT software
Starting a few months ago, I was employed to migrate the software entirely to Linux. For this, I have myself and one other person full time and a number of support people ordering and building kit. We started out by designing a scalable database back end and then we rewrote all the dynamic NT scripts in Perl. We are using Apache + mod_perl as the platform to run it on. Rewriting the scripts in Perl took two months. We are now spending a month in a closed beta test of the new system, and expect to spend another month importing all the users and their data across from the old system. Four months in total.
On the hardware side, it came as a pleasant surprise to find that we could get away with relatively low-end hardware on the Apache web servers. We are currently specifying single Intel P-II 450 systems with 512 MB RAM and 8 GB drives. Our scalability tests indicate that each of these machines can handle the load from around 100,000 subscribers (that's subscribers, not active concurrent users).
The expensive bit is the database. The nature of the web application we have developed is that it has enormous data storage requirements. We archive and store a lot of information on behalf of each of our subscribers. Because there is no high quality database solution for Linux yet - by which I mean no database which offers parallel servers and failover - we are using a popular commercial database on a commercial Unix. This is by far the most expensive part of the system. It looks like we will be paying the best part of $200,000 just for the database / 1 TB storage / back up solution.
Even though we are based in Europe, we have not had any trouble employing bright Linux experts straight out of university. Linux users tend to be a lot smarter than NT users too, since (at present) most people presenting themselves for jobs as `Linux experts' have had to take the initiative and strike out on their own. They are usually more motivated too for the same reason.
Some quick lessons learned: (1) You must make sure you employ people familiar with Linux. (2) It helps if management are behind you on this. Where I work, management have stated that they never want to see any more NT systems deployed. (3) The biggest technical problem to adopting the new system was converting all the old data (in NT proprietary formats) into open formats. If you have a lot of data, then this is a very important cost to factor in. Luckily if we decided in the future to move from Linux to another platform, then we would not have to pay this cost again.
IMHO, you are rolling several very different questions into one. That is, as network service models to be analyzed,
*nix vs. {NT, Netware}
Linux vs. {NT, Netware}
NT vs. Netware
Linux vs. NT
Linux vs. Netware
cross many domains and will yield different answers depending on the problem to be solved and the way the questions are phrased.
In my experience, for example, NT has about 3x the resource cost of Netware in a large departmental environment (200-400 corporate users in 1-2 locations, 20% mobile users). But administration costs for NT go up exponentially as the number of sites and mobile users, while Netware administration costs go up linearly with the same variables.
So if you are trying to convince me to replace NT with Linux, you need to demonstrate that in the equation c = n ** p (c = cost, n = number of sites), p is
But if you are trying to replace Netware with Linux, you are going to have to prove that p
Others have made the argument about NDS, so I won't beat that one to death. But please keep in mind that if you haven't administered a large corporate network, you shouldn't knock NDS until you have used it.
sPh
I don't really think you'll gain that much by replacing Novell with Linux. Now, I'd cheerfully swap out NT for Linux all day, but I can't imagine that you'd get anything but headaches in swapping out Novell.
:)
Caveat: I haven't used Novell since the 4.11 release, as I moved off into NT administration after that. And I have not used a Novell TCP/IP network yet. However, I serviced A LOT of 3.12 servers and a few 4.11.
The reason I like Linux so well is the same reason I like Novell: it works. It works well. It doesn't crash. Novell is just as reliable as Linux, in my experience, and maybe more so. Generally speaking, the only way a Novell server is going to fall over is if the hardware dies.
Now, I have seen unstable Netware installations, but invariably they have been hack jobs by administrators that didn't know what they were doing. A properly administered Netware network is granite solid.
In my mind, Linux's primary advantages over NT are reliability and open source. If something is wrong, I can always fix it with Linux if I'm willing to put the time in to really understand the problem, and once I fix it, it will stay fixed. However, in general, Novell is easier to figure out and it also stays fixed. You do give up a lot of control of the system, but in general Netware works the way Microsoft wishes NT did -- most of the time you don't need bit-level control because it's not going to break.
Novell also thinks their problem solutions out thoroughly. Their print queue is a good example of this. NT sort of has some of the same functionality, but it's a lot easier to maintain complex print structures in Novell and it's a hell of a lot easier to route around problems. Linux's print queues are postively archaic by comparison. You more or less have to build one by hand, and magic filters are a pain in the butt to figure out. Now, you can do more with Linux and print queues: once you understand it you can do just about any neat hack you like. But for most of us, Novell's approach is better. You can start printing with a minimum of fuss and go from there. Linux makes you jump through friggin' hoops to get everything configured properly. That's true of a lot of solutions in Linux.
Novell requires technical skill but has excellent documentation, is much more consistent, and is built with the big picture in mind. You can make HUGE networks with Novell. Most Linux apps aren't designed with the same kind of scalability in mind. They are often written by people with experience with small networks who have an itch. There aren't nearly as many working with really big nets, so there aren't nearly as many good solutions to the BIG problems. Novell is in the business of charging lots of money for solutions to the BIG problems, and they provide excellent, thoroughly-thought-out ones, too.
In my opinion, you'd be best off leaving your network core on Novell. Use Linux, but use it alongside, not in replacement. If you need the whizbang TCP/IP capabilities, or want to deploy a free web server, then you can plop down Linux boxes wherever you like. And when NDS for Linux ships, you'll be able to glue the systems together very closely with a minimum of fuss (apparently).
This is the strategy we have taken. We don't mind having Samba and NT next to each other. (we have no Novell in my present job, which is rather a shame.) We try to run most Internet-related protocols on Linux(we'd use BSD but I haven't learned it yet), and run Microsoft stuff on NT, and everyone stays happy. Samba interoperates nicely in a network with a PDC already in it.
There just is no sense in throwing away your existing investment unless it is too costly to maintain. Linux will certainly help save on maintenance costs once you grok it, but the upfront learning costs are steep. Amortize those by doing little pilot projects until you really know what you're doing. Someday, you may wake up and find that Linux is the center of your network -- but you also might not. And that's okay.
Remember, Open Source isn't going away. It's here for the long haul. You don't have to make this an either/or battle. It's not NT OR Linux, it's NT AND Linux. Take advantage of Open Source where it can help you, and try to give back and make it better. If you insist on making it a war, someone has to lose, and it could be you. If there is no war, then there can be no loser.
-- Ron
Funny to see this question growing up to the level of Slashdot. We had exactly this situation two monthes ago. Yesterday we scrapped the last NT box. Now we are working in a Linux/Netware/Solaris/AIX environment.
;)
Imediate costs and benefits? Well implementing Linux is a tremendous headache. But results overcome expectations.
We are a group divided by three/four teams with 30 people around and serving a whole University network. We had a very interesting place around here called "Internet classes". Some sort of Internet Cafe without coffee. It consists of 70 workstations, several servers of different types and serving more than 2500 users of a wide range of professions and knowledge. The stuff here is quite serious. It is mainly IBM and Sun hardware and it costs a Hell of money.
We started the whole thing on a Solaris/Novell/AIX/NT environment. It didn't work. For 6 monthes we were turned to janitoring constantly a system on which NT was giving tremendous headaches. Things became a major crisis after a serious crash that stopped work for nearly a week.
Some of us lobbied to see NT replaced. After the NT Armageddon's day we got green light to deliver something on Linux in a month. It was Hell. We had problems and problems and problems. But we came with something two months ago.
Results? It is working. Last week we destroyed the NT image that we kept in any case. The last NT box had oxygen cut yesterday at 16:30. People are facing a system several orders more complex than before and are not deeply concerned about it. Some even love it.
The fearful "We want NT back!" didn't last more than a week. Crashes, hangups, coredumps happen but it is several orders of magnitude less than before (some of these new boxes worked nearly two monthes round'clock. A powercut spoiled things).
Do people fear such primitevenesses like the commandline? Yes. But soon they will start (re)using it because there are things that do their job much better than their desktop brothers.
Before we were forced to stick users to "one desktop, one environment". Right now we have a mega-crazyness here where each one chooses the working environment he likes most. And changes the environment as he likes. Before we had a cracking epidemics here. Now we have hackers digging up deeply on the system and helping things grow up. Before we forced everyone to accept the administrative fascism of having a restricted set of apps. Now security is based mostly in supervision. We allow people to do almost everything with their systems as users of a classic UNIX environment.
Right now we have not only a reliable network where people have finally the chance to work. Right now things are overcoming expectations. These two classes are no longer "cafe's". It is Linux Heavy Industries. It is a place where Internet lives not only of browsing. Some users turn their environments to full-fetched working horses. Others use it on developing apps for Internet. And all of us have seen a tremendous boost on development overall. We now are growing things and not janitoring like before. Meanwhile teams are getting free to care for areas that were not touched during the NT hiatus.
Besides we are now planning the future. We are trying to implement things such as clusters here. As an experiment we joined the RC5 contest. Amazingly we are getting a good place over there. There are plans to implement Coda or a huge webcacher based on Squid.
Want to turn to Linux from NT? Take attention to these permises:
You have to take some good time on tuning your hardware. Even in "smooth" installation may hide a lot of "features". And a good careful tuning may turn a rather "slow" machine into something amazing.
Check all those apps over there. Linux is still quite loosen on configurations. And that's a HUGE +. You have a very good chance to fit things much more to your requests.
Office apps are a big problem in Linux. But our world doesn't end here. Try to find a middleterm. When users will note everything else, they will stop whinning and start to hunt you for even more.
If your main language is not english you may find a problem. But most of it is pure cosmetcis. The final result is more rewarding than NT.
Linux is not beautiful. Because power means rawness. And its up to the user to paint it. When he gets there don't get scared. Everyone has a Picasso inside.
Linux crashes. Programs crash. However you have a wide field to maneuver. Even dealing with commercial "crazy" apps is much easier and there's a bigger chance to find solutions.
Take time. The most important is to make a good pre-installation. Get everything you have and try to make the "box of your and their dreams". Check and test it carefully. And then give your users a "gift".
They will cry over you. They will ask your head. They will pass near you and imagine that see you with a rope on your neck they'l be willing to push. But it will not last long. If things were made carefully and well-planned, then NT will be History in a few weeks.
Get a small development team around you. That's very important. Linux is growable and fetching things to specific needs gives a powerful boost on work. If there are no chances to get a developer then roam Internet as much as possible. Even a small sh script may improve things a lot.
Beware. In a few monthes you may note that you are no longer a network administrator...
Actually, it is easy (no sneering :) I tried to figure it out months ago, it's not hard, but the problem is that there's just no step by step procedure outlines. So I made one. See http://socrates.mps.ohio-state.edu/~ccunning/samba .html. That simple procedure is all it took for me, can't guarantee results for everyone :) One word of caution though, access restrictions aren't really well implemented yet.
This is interesting,
Most of the time I see cost savings in terms of
how much can I save by not moving to NT from a
costly Unix. It is more difficult when you can't
use the 40K per box savings argument. One place I
see as a major savings is in applications support.
I currently support 400+ Design tools on several
Unix platforms, my peers who are supporting
similar tools on NT are not even close. NT likes
things to be local, it likes things to make
registry changes instead of environment vars, it
likes dll's in places they shouldn't be and most
applications refuse to be a 'Network only'
install. The more applications and the more people
you support the worse the problem gets. I can
install a Unix tool for 1 person or 1000 at the
same cost. NT tools that want to touch the
registry require overhead for EVERY customer.
-SOTTEK
As far as I can tell, there aren't too many situations where NT or NetWare or Linux/Samba is going to offer much more hardware bang for the buck than the others. Differences of 10-20% don't mean much until you get into fairly large servers.
But one thing I've noticed is that when we buy an NT server, we always end up specifying a bigger one than we really need, if for no other reason than to make so the installation and reboots go a little faster. I'm not talking about crashes, just routine reboots that we're inevitably forced to do for minor configuration changes. When your servers are lightly loaded, as ours are, there can be actual cost savings using Linux.
But the hardware costs for such servers are usually less than the cost of the people who manage them. That can cut both ways. On the one hand, there are plenty of people who can keep a simple NT network up and running, and it's harder to find people with Linux experience. But one of the best-kept secrets about Linux is how incredibly easy it is for a competent person to manage. My NT experience vastly outweighs my Linux experience, but when an NT server gets cranky, I still get cold chills. On a Linux box, I calmly look at the log files, and usually find the answer pretty quickly. Samba's SWAT web admin tool is killer, far easier to use than anything in NT.
So, it all comes down to people. If you have people who are sharp, who understand what's really going on while they're clicking "Next... Next... Finish," then they should be able to do a lot more neat things in a lot less time with Linux.
Much also depends on how the company accounts for costs. Smaller businesses often ignore support and admin costs ("We have 3 people in our IS department, and they're on salary, so nothing's really going to change our costs.").
One other consideration, is that it's often useful to think not in terms of migrating but of integrating. It's relatively easy to pop a Samba server into an existing NT network. Keep doing that, and eventually all you'll have on NT is a PDC. So think of file services and authentication services separately. Linux can be used as the workhorse file-spitter-outer, while you maintain some other system, whether it be NIS or NT or, in the future, NDS, as the authentication system and user database. (hmmm... couldn't Samba use PAM to authenticate against a Novell server?)
I think it may be too premature to look at migrating from NT to Linux right now. What is needed is concentration on Linux-NT coexistance and working together. Linux will have to put the foot in the door by getting in as Web Servers, Proxy Server and File and Print Server. Co-existing with NT. Over time as SysAdmins get used to Linux and also understand that Linux is better than NT they will shift to Linux by pointing out the robustness and cost effectiveness of Linux. This will also earn them brownie points for reducing the cost of IT in the organisation. It should be their idea.
I work in an office where NT and Linux coexist. We plan to use Linux as a proxy server. We are also getting a linux based web server. We are not moving desktops to Linux yet. But you can bet that over time we will slowly but surely.
O this learning! What a thing it is - William Shakespeare