Link to home
Start Free TrialLog in
Avatar of write2safari
write2safari

asked on

why linux?

dear experts, why Linux instead of Unix?
any reason 4 migration from unix(lastest version) to linux(lastest version)?
what are their differences?
Avatar of psimation
psimation
Flag of South Africa image

1)Price, Linux = $0, Unix =$$$
2)Availability of source code for nearly everything you encounter.
3)Very large Open Source community that provides help and advice.
4) Very large development community as well, making the appearance of new features very rapid, and that includes bug fixes and security updates/patches...

BUT, in the end, it all boils down to your needs...
Avatar of ryanf
ryanf

For me...

#1. Community support...

If I have a problem with something in Linux I have almost countless avenues of support... with Solaris or HP I need buy a contract to ask questions about software, and another for hardware...

Not all Unix costs money, I can DL solaris from sun... but you still get stuck buying their support...

#2. Hardware compatability

I run Linux on Alpha, Sun, Intel, S/390, and even seen it run on HP-9000 RISC machines... They all look the same, and same great support :)

just my $.02... hope that sheds some light...

-Ryan
If you already have the hardware and software, then it strikes me that it will be very very difficult to argue for switching OS's from Unix to Linux, as this will involve a major effort.

If you're changing other hardware/software anyway, then you might rethink the whole thing from scratch.
Sometimes price/performance is a reason to switch to Linux.  

I worked for four years at a large company with a big research arm.  We mainly used Sun and SGI in the server room, but when it came time to install a new cluster for some very intensive number crunching, it turned out that we could install a Linux Beowulf cluster for a small fraction of the price of any proprietary Unix.  After I left, I heard that they ended up migrating about a quarter of the server room to Linux over time, as machines were retired and consolidated.

It depends on what the Unix server is running of course.  Not all server-side apps are certified to run on Linux yet.  There will be very few banks running their business entirely on Linux.  But the list of apps is growing very quickly.  It is certainly taking a big bite out of the small-to-medium Unix and Windows server market.

Some companies find that they can buy one or two huge, fault-tolerant boxes, put Linux, and consolidate many hundreds of smaller machines.  Telstra did that a few years ago, replacing a whole room full of Sun boxes with a single large IBM mainframe, hosting hundreds of VMs running Linux.

Last time I checked, IBM was selling 10% of its new mainframes with only Linux installed.
no reason I can see if you have recently installed "latest version" of Unix.  RedHat/Mandrake/SUSE are not IBM.

are you also an expert in Unix?
IBM is a reseller of RedHat, Mandrake and SuSE, and ships a shed-load of SuSE Linux on new (and some old) mainframes as well as x86 servers and midrange.  You can get 24x7 IBM support with Linux, but still retain all the benefits of the giant Linux developer community, plus of course full visibility of the OS code.

Then if your hardware needs change down the road, you don't need to dump or port all your Linux apps, they will usually run on the new hardware either unchanged or only need a simple re-compile.

Everybody wins this way: IBM cuts their R&D costs on software, customer doesn't get so locked in, IBM can let their hardware and support provide the value.  A lot of CIOs like the sound of this.  
IBM invested a big chunk of money in SuSE last year when they were struggling.  SuSE seems to be getting back on track now, I'm glad to see.

(NB: No, I don't work for IBM. :-)

But as I mentioned above, *sometimes* price/performance is a reason to switch to Linux.  That implies "sometimes not".  Linux still lacks some useful high-end features of proprietary Unix, and sometimes these are more important to a certain business application than price/performance. Some of these include:
- Unix's superior SMP scaling,
- High-end clustering
- High-end distributed system mgt
- Large disk, storage array and SAN support
- some mission-critical apps like CICS and VTAM (last time I checked!)
- Journaling file-sys support (I know, it's available now on Linux, but it's still somewhat immature)
- etc.

But you don't always need those things on all your servers (e.g. web and file servers, firewalls and proxies, departmental database servers), and I think in my lifetime Linux will catch up to Unix on all of the above.  Running Linux on mainframe VMs already gives you access to a few of these features.

So Linux server-side is currently a compromise that sometimes makes sense, and you need to look carefully at the application requirements on a case-by-case basis.

write2safari, were you asking in a general sense, or do you have a specific application in mind that you're thinking about moving over to Linux?
I've read a lot of arguments for using Linux. And I believe that Linux is a pretty good choice for many applications.

But, I've heard no arguments for _switching_ to Linux, assuming write2safari already owns the hardware/software he needs (which he/she hasn't clarified).
I 2nd chris - in particular, if you are a current *nix corporate user of IBM/Sun scale, why should you switch to Linux?  Maybe migrate to, or use it for new  non-mission critical applications (if you are willing to become a multi-OS shop).

From a user/developer perspective, what does Linux offer over other *nix? In particlar, what is it's advantage to the owner  in total cost: hardware, OS software, application software, hardware/software maintenance, administrative staffing and time-to-get that critical problem fixed?

Open source can be a burden as well as a blessing. Do  you really want to maintain all your system software? What happens 3 years from now when you want to migrate to "next generation" Linux?

1 of the responses referred to economy of scale - wouldn't this be true on any *nix?
Well, Linux is actually really more robust than the other wintel distos of Unix because so many more people use it and support it. You really have more with linux like the login screens avail. to the CDBurning - games.  But for business uses you may want to use somthing like Solaris v8-9 for certian things. Linux is also easier to operate which helps too!

go with Linux my friend..  
As regards total cost of ownership, again we can't generalise about that, it must be looked at case-by-case.  In the Beowulf cluster example I cited above, we were being asked to pay a high 6-figure sum for the equivalent solution on Sun Sparc or SGI, versus about 50k for the Linux solution.  Because we already had a decently skilled IT dept in-house, time-to-fix was thought by mgt to be same or better, and the savings were more than enough to buy contract support in as needed.  We never needed any: it just ran fine from day 1, following about a week's work on learning how the clustering worked.  Saved us a fortune from the first year deployed.  But we also used a highly specialised, proprietary chemo-informatics application that was only ran on SGI Irix: it made no financial sense to attempt to re-write that, and it's still there.

I've often found that time-to-fix on Linux hardware is faster than with Sun, IBM or HP: I've been kept waiting for fixes by overstretched support groups, especially in offices outside the US.  Not always, but often.
 
This is also true with software: in my current job, we have standardised on Oracle for database and J2EE apps server.  New Oracle products often arrive with unusably bad or missing documentation, and it can take ages for fault resolutions and fixes to come through, and we can't afford their premium support costs.  
So we are considering migrating some of our apps over to MySQL, Postgres and EJBoss: low-end apps that don't require robust transactional support are OK on MySQL, medium-size in-house apps can run well on PostgresQL, and we can leave our financials on Oracle until we're confident of an Open Source alternative.  EJBoss gives us more flexibility that the Oracle AS, since they have no commercial interest in locking us into a single brand of database.  

All of the above can be supported outsourced if we like: we don't need to maintain anything ourselves if we don't want to.  But the key there is "if we don't want to": we should be going outside for support only because we want to, not because we have to.  

I like to fix my own car myself: my brother prefers to send his car to a garage.  "Proprietary software is like a car with the hood welded shut."  In business it's not only a matter of preference, it can be a matter of reducing critical business risks.
I would guess jdfox works at a medium/large business organization - 1 that has the resources to support 4 different databases. I have worked at large financial orgs. that, I believe, would be very reluctant to  invest in support beyond their Unix/Sun/Sybase and VMS/IBM/Db2.

If you are a business using "wintel" (more acurately _ intel)-class Unix, you are most likely small to medium and  will find it hard to cost justify porting/converting to Linux. I am sure there are "important" *nix applications that run on Sun/IBM/HP which are not available on Linux.

Another approach to mixing *nix is develop on Linux, run on Unix.  I had a project where I locally developed a C/Perl/Mque application on Linux and, with only a few #ifdef's, it recompiled and ran on a remote AIX.  I am not sure about version, but Oracle and Sybase run on Linux.

We don't plan to support 4 dbs, we plan to support 3 in the short term, 2 in the medium term.  We already support 2, like all but the smallest organisations.  Our plan is to migrate low-end DBs from MS Access over to MySQL, and then the high-end ones over to PostgreSQL if our eval is successful, with financials apps following slowly and carefully. If Postgres is not up to the job now, then we will wait.  
But regardless of what happens on apps, we are committed to migrating all our servers to Linux, as in all instances we've proven significal annual cost savings on all hosts from year one of migration.  Interestingly, we have one Sun server which can't be upgraded to Solaris 9, but runs fine with SuSE!

JBoss is not a database, it's a J2EE apps server:
http://www.jboss.org/

Many large financial orgs have more recently found that putting all your eggs in one basket is not the best value or lowest risk:

http://www.wired.com/news/infostructure/0,1377,57503,00.html
http://www.forbes.com/home/2002/03/27/0327linux.html

Linux can reduce your risks, and save you a pile of money.  I can remember all the same Fear Uncertainty and Doubt criticisms levelled at Unix by mainframe people, at one time.  "You can't possibly run a large business on Unix" was their tune back then.  The truth is that some things are a better business proposition on Unix, and some are still best left on a mainframe.  Now Linux has come along, and we find some apps are better off hosted on that.  This is why IBM, Sun, HP etc. are now investing heavily in Linux, porting high-end features to it, optimising it for their hardware, etc.: they can see which way the wind is blowing.

It's impossible to back up an assertion like "you...will find it hard to cost justify porting/converting to Linux."  The only way to know is to look at the numbers case-by-case.  
lest we forget, write2safari original question was:

  "any reason 4 migration from unix(lastest version) to linux(lastest version)? what are their differences?"

The question is Unix->Linux.

we have no feedback from write2safari(w2s), but I assume w2s is in a small to medium business. jdfox describes a lengthy,  probably costly, migration in what I assume is a medium to large business.

Running the "lastest version" of Unix implies, to me, recent upgrading (altho this might be hypothetical)

I also believe 'there are "important" *nix applications that run on Sun/IBM/HP Unix/AIX which are not available on Linux.'

As jdfox point out, there are currently issues of "not up to the job" on certain Linux applications - as well as mission-critical issues.  

Considering this, a small to medium business "will find it hard to cost justify porting/converting to Linux."







Why assume that w2s is in small to medium, since w2s hasn't said?
Why assume that it would be lengthy or costly, since w2s hasn't said what his apps are?
Why assume that he has "important" apps not available on Linux, since w2s hasn't said what his apps are?
Why assume that Linux won't be up to the job, since w2s hasn't said whether he/she currently uses or plans to use those features?

Considering this, you have no idea whether it will be hard to cost-justify or not.  Why not wait and find out what the facts are, before coming to a conclusion?
OK, jdfox how about your organization?
  1. small,medium, large?
      1a. size of IT staff?
  2. project duration (from feasiblity study through cutover of major apps)?
  3. staff-months to implement above - including user staffing for planning & testing
   4. I said as you said: "Not all server-side apps are certified to run on Linux yet."
      I should add -there are some apps which only run on a certain flavor of Linux.
   5. BTW, do you know of any business apps that are unique to Linux?
   6. You say "(as) your hardware needs change down the road, you don't need to dump or port all your Linux apps, they will usually run on the new hardware either unchanged or only need a simple re-compile."
      Can you quantify "usually run"?
   7. How confident are you that if your platform requires a software upgrade, either/and :
     A  New application version
     B  Kernel/lib

     - that your apps will "usually run"?
Lest we forget, this is about write2safari, not me, and not you.

1. Irrelevant.
2. Irrelevant.
3. Irrelevant.
4. True.  But until we know what apps he's running, irrelevant.
5. No.  But until we know what apps he's running, irrelevant.
6. No, and can you quantify "lengthy", "costly", etc.?  Of course not.
7. It depends on the apps.  I've never been completely stuck porting from Linux to another, even after major glibc and kernel changes, but I've been completely stuck porting from one Unix to another, since I'm often reliant on one vendor's tools or libraries.  
it is clearly not about me since I have used Linux for a long time but only as my personal/home system. I have used AT&T,Sun & AIX for even longer, but never converted OS's except as noted above (AIX/Linux cross-compiling).

you, on the other hand, appear to be the only respondant who has actually gone through such a conversion(s) and are familiar with "price/performance is a reason to switch to Linux"

Your numbers would be of interest to write2safari and others who might be considering it - and to me.

As ballpark, I would quantify a Unix->Linux conversion as "lengthy" if it took over 4 months including cutover and testing. Costly if I did not recover my net costs after the 1st year of operation.  What would be your ballpark?

BTW, if we did have a list of his apps, how might we know the technical feasibility and length? You give the example of Access to MySQL. Can this be done as a "bulk/one-shot" conversion or is it per Access-app?
Hey guys, Isn't Solaris for Wintel machines free?
arn0ld, thanks for your polite reply.  Fair questions, though we'll have to guess about write2safari's interest, since he appears to have unsubscribed from this thread. :)  

By way of intro, I work nowadays at a small, badly underfunded university, with a small but highly dedicated staff of about 20.  I guess that makes us "small-to-medium", depending on who you ask.  We are all pretty flexible, and happy to learn new technologies.  The job gives us the freedom to innovate and "build-out", albeit for appallingly low pay.  We are extremely budget-driven on every little damn thing, and being public sector, we can neither boost revenue by raising prices nor go to the stock market.  Our revenues are fixed, but our costs rise.  So we in the IT group constantly hunt for ways to boost reliability and cut overheads.
This outfit is historically an Oracle and Sun shop, with a natural distrust of all things MS, as we've seen so many other universities burnt by them.

On the Access to MySQL side, I can give precise details, though that's not a Unix to Linux conversion, it's a Windows to Linux conversion.  
We have a huge number of Access apps littering the place, mostly poorly designed and badly maintained.  Some have wound up holding business critical data, which is a real risk.  We looked at a large number of schemas, and found they were all either flat file or relatively simple relations between tables.  Rarely are they used by more than 2 people at a time, so no real transactional issues, rather, they need speed, and above all simplicity of maintenance and change.  MySQL is perfect for that sort of thing, and the new version even has a decent transactional capability, though it is quite new and immature.  I think it will improve rapidly over time, as it has a huge installed base and a very active community working on it.  We are working on giving the users ODBC access to a MySQL farm from MS Query now, and over time we hope to replace MS Office with OpenOffice, and gradually get rid of Windows altogether.
Conversion was about 60% one-shot, 30% per-app, and 10% complete re-write for the really hopeless amateur Access jobs.  We will have finished 3 months after starting.

Linux was introduced here by a small band of enthusiasts.  Old story.  Mgt were at first worried for the same sort of reasons you outlined, so we documented the proposed architecture.  They then said "cost it out in detail", and we did.  When Sun announced their Linux strategy, that finally nailed Solaris in mgt's eyes. We have a Solaris 7 server that's been giving us no end of little charges here and there for the past year or two, and an equivalent horsepower Linux server that's been running without a penny spent on it for 2 years.  This year the cuts have been particularly fierce, so mgt said "go for it, and keep the Solaris on standby in case of incipient doom".  We got a slight performance gain, for no extra cost, running Oracle and several busy apps, with about 2000 users.  One vendor's proprietary app running on there had deprecated Solaris 7, so we tried to upgrade it to 9.  Couldn't do it, and Sun wouldn't ship us a patch: we had to upgrade the entire SCSI array and backplane first!  SuSE ran fine on the existing hardware: that also impressed mgt.  But they liked the idea of hosting on x86, so over we went.  The PC cost about 2000 dollars, roughly the cost of the SCSI hardware.  We would also probably have spent money on Solaris 9 training, and other bits of hardware and software, going by history, so it appears to have been a net saving in same quarter.  We've frozen the Solaris server at 7 now.
We also have a number of older Sun servers that are nearing end of life, and Sun are pitching a big fault-tolerant box at us for server consolidation.  But we could get better performance and uptime with a rack of 2- and 4-processor Linux Intel boxes.  We'll be saving about 30 thousand in hardware over the lifetime of the consolidation, probably about two years.

As for his apps, well, let me try to pick some examples:

If it's an Oracle/DB2/etc app, moving it from Unix to Linux will almost certainly be with code unchanged or minimally tweaked on the shell scripts, unless, for instance, there's significant amounts of external C routines running outside the db.  

No code change at all if it's a J2EE app on WebLogic or WebSphere, only moving OS, and not app server.

Minimal changes needed moveing from one Java app server to another (e.g. WebLogic to JBoss).  An experience EJB hacker should manage it in a few weeks, using good tools like TogetherSoft.

Moving persistent data store from one database to another (e.g. Oracle to Postgres) would mean a fair amount of SQL tweaking here and there, especially if you use a lot of those nice lock-you-in features that Oracle love to push.

A custom C/C++ app would probably mean lots of porting work, going by my own past experience.  I'm pleased to say we do almost zero development in C/C++ anymore, nearly all is high level languages like Java and Python.

The conversion time including cutover and test, well that just varies too much to say.  It depends not just on your apps, but the number of man-hours available to you, the skill of the staff, etc.  Give me a precise example of something you run on Unix, and I'll tell you if we could port to Linux ourselves, and how long it would take our team.
thanks for the enlightening and interesting details.

your Linux successes wood seem to bode poorly for Sun.

This makes it anomolous to me that if you put "linux" into a Dice (I assume they have an ad on your web page) search, there are very few matches and even less if you remove IBM Linux.
I'd never heard of Dice.com before, thanks, I'll bookmark that.  You're right, there are fewer Linux jobs than Unix jobs on there.

There's certainly a lot more installed Unix around than Linux, and these hosts won't be replaced with Linux before the machines reach end-of-life.  Beyond that, it depends on how soon Linux can gain the enterprise features we discussed.  3 years?  10?  Never? People in their 40s can rely on Unix being around 'til the end of their working life for sure, but the good thing about Linux is that it's functionally so similar that any good Unix developer or admin can easily add it to their skill set. Unlike some other OSes... (*cough* windows *cough*)

It's famously hard to say how much Unix (and Windows) are being replaced with Linux: does consolidating three old Sparcs into one big Intel Linux host count as "three hosts replaced" or "one new host"? Does Linux replacing Solaris on Sparc get counted, or does the market analyst only count Linux on Intel?

Here's an interesting article by Nick Petreley about the numbers game on market share:
http://interviews.slashdot.org/interviews/03/03/24/1425201.shtml

Sun can re-invent themselves with Linux long-term, though it will be hard work for them.  I read this morning that they've just dropped their in-house distro, and are going with Red Hat instead:
http://www.eweek.com/article2/0,3959,981353,00.asp
http://www.informationweek.com/news/IWK20030331S0009

Does this mean they'll buy Red Hat?  Red Hat stopped supporting Sparc last year, so does this mean Sun will pick that project up, or are they moving slowly away from Sparc?
Weird times...
write2safari:
This old question needs to be finalized -- accept an answer, split points, or get a refund.  For information on your options, please click here-> http:/help/closing.jsp#1 
EXPERTS:
Post your closing recommendations!  No comment means you don't care.
No comment has been added lately, so it's time to clean up this TA.
I will leave a recommendation in the Cleanup topic area that this question is:
PAQ / No Refund ; 20 points can't be split........
Please leave any comments here within the next seven days.

PLEASE DO NOT ACCEPT THIS COMMENT AS AN ANSWER!

TheWeakestLink
EE Cleanup Volunteer
ASKER CERTIFIED SOLUTION
Avatar of YensidMod
YensidMod

Link to home
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
Start Free Trial