Abandoning HPUX

[Jeroen Peereboom]

Mark,

of course you can't replace all you HP-UX boxes because of the applications running on them.

But some of them may be replaced by Linux.
I'm currently involved in an Oracle 9i RAC Linux cluster and I sense the installation is a bit more problematic than expected. We didn't get support issues arranged before starting.

Problem with Linux is you cannot afford not to apply the latest (security) patches. It takes some effort to keep the servers patched up-to-date.

If you buy Intel based servers you are more flexible how these servers will be used. You have one hardware platform available for Linux and Windows (I suppose you have Windows servers too). That means less effort for hardware maintenance.

Linux can run on PARISC too?!

JP.

[Stefan Farrelly]

Very interesting. Weve been playing with Linux to replace our HP's (Were mostly Oracle) for some time now and Linux as a platform with Oracle 10G is almost there (once 10G is released later this year).

For now an Oracle license costs as much on HP as Linux so no incentive there to spend a huge effort deleloping a new platform - considering how nice and stable our HP/Oracle platform is.

However, costs of HP hardware is a big difference to an equivalent Intel machine running Linux. Weve just bought 2x Superdomes running 8x HP partitions (no one with >8GB RAM and 8 cpu's) and you can do all that on Intel for a fraction of the hardware cost and maintenance cost that we paid. Big incentive to switch there - not for current projects, but new projects and as we upgrade old projects.

Software costs is not an issue - HPUX is very cheap, and Linux is free, but any decent Linux you have to pay for too. MC/SG is available on Linux now too for those mission critical apps (if you dont like AS).

The SAN setup is comparable on Linux as it on HP and we already run Linux on our EVA's and EMC's via SAN with same throughput as on HP.

Once Oracle 10G is out we will test large Linux clusters on it and im confident it will perform as well, and be almost as reliable as HPUX, for a large reduction in hardware and maintenance costs (for us a saving in the hundreds of 000's per year).

Good luck!

[Stefan Farrelly]

Another point - dont forget Amazon is running on a massive Linux cluster run by HP - and it performs pretty good - or else Amazon would dump it.

[Nicolas Dumeige]

To my point of view, the purpose and usage must dictate IT infrastructure. On numerous domain (firewall, desktop, ...) Linux but also all free BSD Unix, have a lot of thing to give.
All blend of commercial Unix have its speciallity, Linux and other are general alternatives. Linux is supposed to be very scalable, from the watch to the mainframe virtual domain.I guess that as the time goes on, we'll see if it can do all thing with efficency and avantgarde spirit.
All commercial Unix are declining in profit of Linux and M$ Windows, pick your rival !

[Jeff Schussele]

Hi Mark,

Well. it's days like this that keep life interesting.

I find myself agreeing w/Tully.
I would think this would be an argument about TCO & not capacity/performance.
I really think they'll be hard-pressed to prove that an Intel box could outperform a properly sized 9000/Integrity. And I think Integrity is the key here. I'd argue that they get a lower end RX system & try their hand at RH Linux before they totally commit to that "strategy".

My 2 cents,
Jeff

[Mark Grant]

Thanks everyone,

This has given me a lot of food for thought. I'm not as against the idea as I was (which makes me happy because I didn't want to be against it) so now I am 50-50.

In a way , I think it is inevitable for us all really, in the long run.

Any more comments gratefully accepted.

[Steven E. Protter]

On vacation. Missed this nugget.

We have not had had any primetime downtime in production attributable to our HP-9000 servers other than human error since 2000.

That has been due to a combination of good systems administration and the right tools. Our 9000 servers are relaible, even the old ones. They have not let us down.

The operating system itself is like a rock, running smoothly, with great tools and redundancy like mirror/ux etc.

I run both types of systems. Linux for my private webhosting business and HP-UX at my job. The obvious advantage of Linux is up front cost. The obvious advantage of HP-UX is scalability, reliability and a the millions of man hours HP has put into making it what it is today.

Linux is great, but there still is an ease of use and reliability gap between it an HP-UX, this has been proven.

For flexibility, I'd recommend Itanium hardware which can run HP-UX, Microsoft OS and Linux. It gives you teh most flexibile tool for the job.

SEP

[Hemanth Gurunath Basrur]

Hello Mark,

Kindly have a look at the foll. links:

http://h10018.www1.hp.com/wwsolutions/linux/index.html

http://h30046.www3.hp.com/search.php?topiccode=linuxnews

http://h30046.www3.hp.com/search.php?topiccode=linuxCASESTUDY

http://www.hp.com/hps/linux/

http://www.hp.com/hpinfo/newsroom/press/2003/030319a.html

HTH.

Regards,
Hemanth

[Martin Johnson]

Remember, the key is the application software that will run on the system.

First, does the application run on the OS? Most of our appliaction do not run on Linux (yet).

Second, how mission critical are your applications? Can they accomodate down time? HPUX is more stable in critical environments.

Does Linux have some advantages over HPUX? Definately. We are migrating our multi-threaded simulation software to a blade GRID server running RedHat Enterprise Linux AS 3.0 and Platform's LSF GRID software.

Bottom line, take a good look at your applications and the direction they are taking.

HTH
Marty

[Todd McDaniel_1]

Mark,

one of the first things I would do is call your HP salesman and tell him your company is thinking of bailing on HP for Linux...

Im sure he will be more than happy to work out a deal!

Hehe...seriously, One of the best things you can do is start a bit of friendly competition.


1) look at your compatibility issues (oracle, etc...)
2) look at your cost for switching (new hardware agreements/and hardware)
3) look at your retraining issues
4) look at what transition period it will take to roll in the new systems.

[Chris Vail]

As you know, this is a management decision, not a technical one. There are good arguments to be made for either choice. I like the idea of a heterogeneous environment, rather than a single vendor.

IBM is making a HUGE move into Linux. You don't even have to buy their hardware to get a support contract from them. If I were you, I'd definitely have my friendly neighborhood IBM rep take me to lunch and give you the lowdown and dirty story of IBM and Linux.

As someone else mentioned, both Compaq and Dell sell excellent Linux based servers, with everything you could want in terms of hardware features. Since HP now owns Compaq, it may even be a pretty good idea to get the story from HP about their Linux initiatives. But any of the three: IBM, Dell, and HP/Compaq, sell excellent hardware, and they are cheaper than an equivalent PA-Risc based system.

Consider also migrating to the Itanium servers. The HP Superdomes with Itanium can run all three OS's: HP-UX, M$, and Linux all on the same physical hardware. This is very, very cool. I'll wager you could get one of these in for free for 90 days for a proof-of-concept.

My main story is: be flexible. HP-Ux is still a great OS. But that doesn't mean its the only OS. Try a proof-of-concept running Oracle on Linux on Intel or Itanium for a year. Compare the TCO over that period of time. Especially see if Oracle patches and revisions come out as fast on Linux as they do on HP-ux.

As always, technology is in a state of flux. I've got a RH 9 system here running on a Dell 2450 P3/850 that is easily as fast as the A500 I have in the same rack. I haven't done any formal speed studies (and haven't the time/energy) but my overall impression of this combination is positive. I would not hesitate to load up Oracle on it and race it.

Another thing to consider: get an Itanium system and try loading Oracle using HP-ux, and then some benchmarks using your exact data. Then dump HP-ux and load Linux, with Oracle on top of that, and run your benchmarks again. I think you'll find that the Linux is at least as fast as HP-ux, and perhaps faster. I would be VERY interested in the the result of your benchmark.

If Linux is as fast, then you need to do a cost analysis. Include things such as you going to training so that you can become as proficient in Linux as you are in HP-ux. Project that cost analysis forward over your anticipated purchases for the next couple of years. At the very worst, the two analysis will be the same. But I suspect that Linux will be a shade cheaper--even after figuring in the training and transition costs.

To be very cool, let IBM and/or HP do the analysis for you. If you're a big enough customer, they'll trip all over themselves in their hurry to give you that data.

But don't dismiss out of hand the idea that you might have to learn something new. Although I'm a CSA, I now have responsibility not only for HP-ux 10.2, 11.0 and 11i systems, but also AIX 4.2 & 5L, Solaris 7, 8, 9, Tru64 5.1, RH Linux 9.0, and several M$ variants: 2000, 2003AS and NT 4.0. Somewhere in here is a SGI Octane machine with Irix that I haven't fired up yet.

I've decided that the day I stop learning is the day I start dying. And so I'm enthusiastic about learning this new stuff.

Yes, its confusing. But they pay me the same whether I'm confused or not. So I work through the confusion, and come out the other side a better sysadmin.

Just yesterday I tried to add a new volume to rootvg on a HPUX 11i system. The term rootvg is an IBM/AIX term, not vg00, which is a HP term. It took me 2-3 tries before I realized my error, typed in the correct argument and continued on. Its just not that big of a deal.



Chris

[Colin Topliss]

Heh, I really should keep my eye on the forums more often during the day.... this is a good discussion.

I'm in the middle of looking at Linux in the workplace. We're a mixed site, with HP, Tru64 and Solaris. We're looking at moving some of the middleware tier to Linux (from Solaris).

I have to admit to being a Linux fan, but my primary job has always been working with HP-UX.

It isn't quite feasible yet to move everything to Linux - it just isn't quite there yet. For example, our main billing system will be staying on the Superdome for quite some time. But there is a place for Linux. Why deploy expensive PA-Risc systems when a cheaper Intel system will do the job. And lets face it, the uptimes on Linux systems are comparable to all of the main-stream systems we're all used to.

The one thing I do like about Linux is the flexibility. I was testing out (unsupported) FC attached storage (FC60 and Emulex) - and managed to get it working with a little tweaking of the source code. Couldn't do that on HP-UX!

Patching is a dream under Linux. You don't have to reboot during the patching process itself.

The thing that I like most is the expanded choice. One of the things I always hated about HP-UX is the horrendous time I have to spend trying to get applications to compile - either that or I have to contend with what has been made available via the porting centre (which is usually a few versions out of date, and sometimes refuses to compile anyway - long story). I rarely get that problem under Linux.

So yes, evaluate Linux, but only suggest it where it is reasonable to do so (so no, don't run your major billing platform on it, and don't expect to run a data-warehouse on a 4 cpu Intel system when it will barely run on a Superdome)! ;-)

Col.

[dirk dierickx]

one word, hue, well actualy three: proof of concept.

try it out on one box, you probably get a test box for a month from HP if you ask for it. run your tests, see how they perform, stability of the app etc. that is the way linux is getting into our data center. as soon as we see the poc is going nowhere we abondon it. until now we did not have many problems.

and the problems we had were related to:
1. bad apps, nothing to do with linux, just a rush port job of the vendor.
2. bad hardware, just stick to hardware that is supported straight out of the box, you _don't_ want to go the 'compile additional module from vendor' route.

[Wim Rombauts]

To my opinion, the absolute main difference betwwen Linux and HP-UX is the fact that Linux can run on very cheap intel boxes.
When you start looking at costs for OS-support, or that cost of software and software support, the hardware cost is mostly not that big of a deal, but still ...

But do you want to run on cheap hardware ? or would you choose to run Linux on a high-end hardware like the HP rx series Itanium servers ? When you do that,almost all advantages of Linux have gone : There is no real cost issue anymore, and you have got to learn Linux, while you already know HP-UX.

[Sanjay Kumar Suri]

Quite an exhaustive debate. More to add:

We have been using Linux primarily for application servers/Web-server/Proxy servers etc.

Mision critical servers & database servers are definitely no-no.

Over a period of time management is cagey on the Linux support front. So we are planning to shift on HP-UX completely.

sks

[Nicolas Dumeige]

I've read yesterday that at the last Intel show, HP annonced it will stop Alpha & PA-Risc processor research and production.

The "Aplha secrets" (smart parallel execution management routines)were given for free to Intel. In fact, people from the Alpha develeppement team are now part of the intel team.

After that HP will then depend totaly on Intel.

The Itanium2, it not compatible with 32 bits application, an not even with Itanium 1. Unix vendor are (were ?) more didicated to customer trust and line compatibility.

The strange thing, my point here, is that Intel / Linux combinaison, which is presented as the killer-standard-must-be-followed solution is somehow a mess.

Replacing all vanilla Unix with Linux, why not, but there in no Linux per se, there's Debian, Red Hat, Suse, Turbo Linux, Lindows, Mandrake, ... distributions. Ok they are supposed to share the same kernel - maintened only by Linus Torvald - but in the fact there're not. All have there specific configuration file, ..., and so on.
Linux should be standard and unify all the Unix users in one open source stable environnement, well is it ? Red Hat has decided to stop giving it's distribution for free, they have change the profit model. Tomorow, the next commercial commercial Unix will be Linux.

[berlene]

http://www.hp.com/hpinfo/newsroom/press/2004/040115a.html

HP unveiled several new Linux reference architectures, including commercial Linux reference architectures based on Oracle Database/9iRAC and BEA WebLogic Server and an open source reference infrastructure architecture based on open-source software from MySQL, JBoss, Apache and OpenLDAP.

The Linux reference architectures are being deployed on industry-standard HP ProLiant servers, to be followed by deployment on HP Integrity servers. These solutions are tailored to enterprise customers migrating from Solaris, AIX and Tru64 to save them time and money.

The solutions help mitigate business risk for customers and improve the support for mission-critical deployments. HP is the first major Linux vendor to offer these types of commercial applications not from a proprietary, single-source; the solutions were created by integrating applications based on open source software and the systems expertise of HP.

just some info I found

Berlene

[Nicolas Dumeige]

Berlene, with all due respect, what you said sound to me like a HP corporate advertisement...

Linux has been an amazing source of innovation, as a result, it is be essence an anarchic world which must be simplify to become a trustfull entreprise solution.

Standard are seldom pacific agreement, there more like the winner prize. The struggle for standard Linux solution is not yet over, and customer will end up with standard but non-comptible solution.

[John Carr_2]

Mark

it would be very interesting to put hpux v linux to two respective salesmen and listen to all there reasonings, recomendations and general sales spiel on the basis you had a very large budget to spend in the next three months.

:-) John.

[berlene]

Well Nichols, noting the "newsroom" and "press" in the URL was a dead giveaway :-) Give me a break, I haven't had my second cup of coffee yet!

Berlene

[Mark Grant]

John,

I quite agree!

Most of the answers posted have been extremely valuable and it is an interesting topic. However, I think, ultimately, there is no clear winner in this discussion. I now tend to think we should give it a go, and see what happens.

However, this is being seen here as a quick fix for some urgent resource problems and I don't really think that is the way it should be done.

There should be a clear methodical testing phase. I don't think I will get the luxury.

[John Carr_2]

Mark

Ifs its an urgent requirement for a fix I personally would walk away from it, you need time to plan and check - whats the likely hood of failure from trying to rush this against sticking with what you have lots of experience with.

I have enjoyed reading all the comments great thread
:- John.

[doug mielke]

After working with Unix(s) since it was a little baby, I have to note that I've never used one more stable than HP/UX. Also, I see some hints lately that the HP claim of Risc abandonment may be subject to change. Is intel really ready for prime time?

And I know that comments like this help little when trying to justify a decision to management.

There are mountains of doc out there, likely originating in Washington State, that trash Linix's TCO. It by definition can not be as stable as a controlled version OS, by a somewhat conservative co. like HP. So upgrades have to be more common, more patching, more apps versions, more training...

[Bruno Ganino]

Hi Mark

I think that Linux is a Good O.S.!
There are two motives:
1) Available Source Free.
2) Difference of marketing.

The main difference between Unix and Linux is that Unix is proprietary binary-only OS.

The problem with a proprietary binary-only OS that is available from several suppliers is that these suppliers have short times marketing pressures to keep whatever innovations they make to the OS.

Over time, these proprietary innovations to each version of the Unix OS cause the various Unixes to be different.
This occurs because Unix vendors don't have access to the source code of competitive innovations, the license Unix vendors use prohibits the use of such innovations (even if everyone involved in Unix wants to use the same innovation).

Linux is different.... if one Linux supplier adopts an innovation that becomes popular in the market, the other Linux vendors will immediately adopt that innovation.
This is possible because they have access to the source code of that innovation and it comes under a license that allows them to use it.

BUT

I think that Linux Not is still mature for Big systems, it is more adapt to systems plus littles.

Another thing that need to keep present is .... always will be free ?

Regards
Bruno

[Thomas Bianco]

US$.02:

My company is doing the same thing. We decided to replace aging HPUX systems with Linux, as we could purchase new hardware for less then the cost of maintenance on the C110s and 720s.

I should caveat that and say that we are replacing HPUX SLOWLY, over a period not less then 3 years. In addition, we are starting with the "low hanging fruit" of standard Unix Services (apache, NFS, NIS etc).