Abandoning HPUX

[Robert Binkhorst]

Hey Mark,

I've been looking into migrating a database away from a HP-UX DR machine, and these were our reasons for staying on HP-UX:
- Using the new rp3440 machines with 1Gb CPU's the performance gain is very nice, no need yet for Itanium, although if cost doesn't matter I'f definitely have a look at these. You need less CPU's which is cheaper.
- These rp34x boxes are VERY cheap to buy compared to DLxxx machines with the same #CPU's/Mem. When you start to include the cost for OS/Support there's not much difference anymore between HP-UX/RedHat (and both are cheaper than Windows in our situation).
- Ease of management. Although all of our Linux production machines have been running like a charm for the past 2 years (middleware/message brokering) they need more attention to manage. Also, management is not as integrated as on HP-UX. The new HP insight manager offers nice hardware monitoring though.

We found that support is no issue. HP will readily support RedHat, and the lifetime is 5 years for the commercial versions.

One of the main considerations that we had is application bound. We have a site license for HP-UX for Sybase. Running this on Linux/Windows would cost us extra in licensing.

HTH,

Robert

[Thomas Schler_1]

Hi people,

we're running HP-UX for somewhat 15 years. Of course, HP-UX 7.xx is something different from actual HP-UX, but, today, it's a quite stable OS running on very good (but expensive) HP hardware. Support from HP also is very good (every time helping when this forum can't). We have 2 K servers, MC/ServiceGuard, Oracle, and numerous own applications, and several HP-UX workstations (and other OSs).

During past years HP-UX workstations and Windows (what's that?) were replaced more and more by Linux clients. During past year, politics changed and the management just wanted to also have our servers replaced by Linux, mainly because of cheapier hardware and software, not because of a "bad" HP-UX OS.

In my opinion, this is a wrong way. Throwing away a good thing, that has proven to be stable, easy and secure to use, and the daily work is done very well on this system, and replacing it by something cheap, not knowing if the cheap system can do the same job at least as good as the actual system, and not knowing where the direction of open source development will point to in three years (will the community of open source developers provide you with sw solutions you need? Will you get support? In three years, are you forced to change a software just because someone outside decided no longer to develop/maintain his software you're using?), can be a very risky way of driving a complex system.

I would not change the system unless I was forced to do so by a bad HP-UX, by bad services provided by HP, or by politics of your own management. In the latter case, decisions are made after answering the question, who is stronger? You, by trying to convince the management by arguments? Or the management, not wanting to here that they are making bad politics and having the power to cancel your job contract?

Changing the roots (OS) of a running and stable system is like re-inventing the wheel. Normally, I would say, don't do that.

[rick jones]

If you do have to consider a migration, a "split the difference" strategy would be to use IPF hardware - the aforementioned rx series of boxes. On them you could run Linux or HP-UX (or Windows, or OpenVMS...) as desired/required.

About 32-bit on IPF. It is indeed possible to have 32-bit applications on IPF. There is the already familiar x86 emulation stuff under Linux and the PA-RISC emulation stuff under HP-UX. Beyond that however, one can have a _native IPF_ 32-bit binary on HP-UX.

[Uwe Zessin]

Hello,
I am not a system administrator, but when I deal with storage I have to deal with the OS, too. I am almost finished with a project where we have replaced old HP storage with 2 EVA3000 storage arrays connected to a 4-node cluster. It is really amazing, but after the HP-UX admin put in the required patches everything went smooth. On my last visit I helped to connect two more EVAs to another 3 new Integrity servers which will form a new cluster. Again, zero problems.

The Linux history (me and my employer came from the DEC/Compaq side to HP) does not look so good. It took ages to get the multipath software (SecurePath) for the storage arrays and there are still some limitations: e.g. cannot scan for new LUNs - a reboot is needed. With typically 4 paths to each LUN we can 'only' connect 32 LUNs, because each path to a LUN is counted against the 128 LUN limit in the kernel.

I also have been told about stability problems with Linux on bigger hardware and the vendor wants to convince the customer to use his own Unix version - sorry, cannot tell any details.

Honestly, I don't think Linux is ready for "Daddy's stuff", yet.

[Mark Travis]

It sounds like your boss asked you to investigate linux -- not to immediately shoehorn it in to your database infrastructure.

My company's a lot smaller than yours by the sounds of things, but we're moving from Solaris/sparc to Linux/intel. Production DBs aren't slated to go in quite yet, but we'll move those eventually after thorough testing. We've standardized on ProLiant DL380s for low end stuff, and will probably go higher up the HP/Intel food chain when we get there. I like the DL380s even though I've only had them for a week.

Anyway, back to your situation. I agree that Linux/Intel isn't quite where Unix/RISC is. There are bugs in silly crap like "sort" on Linux (RH9, anyway, for example). So yeah, Linux won't bring the Nth level of stability that you currently have. I don't think anybody would disagree with that.

So what I'd suggest is that you find out from your boss what the requirements are. Don't do it from a FUD standpoint -- just say something like, "Hey, we can do with Linux pretty much what we do now and a spitball estimate is that we'll probably save money. But, it's probably not going to be as stable as what we currently have. So what are the requirements? I can't give you identical stability, but can come pretty close, blah blah blah..." That should open up a discussion.

At the very least you should go with RH enterprise or SuSE enterprise. All of the usual suspects (oracle, HP, EMC, bea, etc.) support those platforms.

Your management doesn't need to be technically competent. But as long as they're reasonable people they you should be able to express your *reasonable* concerns before a final decision is made. But if you phrase it from a standpoint of terror on your part then it probably won't help you much.

If the cause of the problem is, as you suggest, lack of planning, then maybe you can find ways to better utilize the existing infrastructure. Like, does every conceivable front end application require a dedicated back end environment? Are you keeping tons of crap online forever whose requirements have long since withered away but nobody has asked about removing?

Anyway, good luck.

[Victor Fridyev]

Hi,

An additional $0.01.
When we talk about LINUX, in most cases we can't have a single support contact point. Let's say, you have RH on IBM and something doesn't want to work. Now you have to find out: where the problem is: in HW or SW area, which means where you can get help from: IBM or RH. This is very critical if you have to install computers with an additional equipment somewhere in the middle of nothing. Here I see great advantage of HPUX: everywhere strong HP support (both HW and SW) exists and it's very reliable.

regards

[Sanjiv Sharma_1]

Hi Mark,

I am not convinced that it is not an April Fool Story but I am very delighted that you have opended such a wonderful discussion where we come to know the cost of migration from HP-UX(or any OS) to Linux.

This has become a hot discussion in the Management nowadays.

All the best.

[Stephen1268]

Quite informative dialog. Thanks for everyone's contribution to get inside the issue of Linux.

[Julio Yamawaki]

Hi,

we have some HP-UX box and some Linux (Red Hat AS 2.1) box. All running Oracle 9i.
I think, for HP and Oracle, Linux will be a "more supportable" platform, as if Oracle test new products in RH AS 2.1, it will be tested in Any platform certified in RH AS 2.1. Of course it will have many more servers with RH AS 2.1 than HP-UX 11i in the future.
Also, for now, I think if you choose to use Integrity servers with Itanium, it is better to use HP-UX (as LINUX has the same price) and HP-UX is easier to manage.

[John Payne_2]

A year or so ago, we got a new CIO out of the CS department on campus. He has been pushing Linux since. The jury is still out.

I personally feel moving from 64-bit to 32-bit and expecting the same performance under load will always be a mistake. Costs to throw more HW and SW at a load problem can quickly skyrocket, especially if you are paying for per-processor licenses (like for Oracle or Peoplesoft) One of the things about Red Hat Enterprise Linux was that they said they would slow down the release cycle. This doesn't seem to be happening yet as other people have said.

Our CIO and other manager types went to IBM and they did their 'magical linux demo' for them. They watched someone magically create a new fully running linux partition on a large machine in a matter of minutes. We figure they were running VM underneath it. We just got rid of our mainframe a few years back and have no plans to go back to it...

We have just finished a 'proof of concept' project for linux, and our online courses system is migrating to linux (32-bit...) this summer. We'll see what happens.

There seems to be less and less of a rush to throw things on linux. One thing we are likely to use it for is to reduce the Mongolian horde (no offense) of Windows servers sitting in the data center. (Maybe throw a bunch under vmware on a bigger Itanium machine.)

I am currently working on an investigation project to verify everything we need runs on HPUX 11.23. If so, our recommendation going forward will be to purchase Itanium machines, run HPUX on them, and be happy. The price for an rx1600 is the same as that of a Compaq dl360. The cost of ownership is no more for HPUX, and may be less these days.

Sure, Linux will work. HPUX will work. Don't be fooled by the idea that Linux is cheaper. It just isn't, necessarily. Just don't tell me you went to AIX. (The Bizarro Unix.)

Hope it helps, and good luck

John