Qantas Ditches Linux for AIX

An anonymous reader writes "Australia's No. 1 airline Qantas will shift their underlying platform running its internal finance systems from Linux to IBM's AIX next month as part of a wide-ranging technology transformation project. 'We're moving from a Linux platform to an IBM AIX environment — we did that to address some stability issues we were having', said Suzanne Young, Qantas group general manager for finance improvement and segmentation. The decision was made last year, as part of the planning for the rollout."

Comments

Re:IBM business plan at work

[snero3]

You are not too far from the truth there.

We started our relationshop with IBM on their intel and Linux X series servers and as we grew they moved us to P series servers running AIX which happens to run all linux binaries just fine and even has the same command set.

The "Upgrade" path was easy and plainless and the cost was spread out over years so it kept management and the accountants happy.

Personally I see it as a winning solution for both Linux and IBM.

Why the surpise? Linux IS NOT the most stable Unix

[keepper]

Ok, put it as experience.. Put it as bias...

But in my experience and that of many others. linux is flexible... fast.. versatile.. but the most stable it isnt.. its part of its design goal. A stable OS, has stable developement practices.. Linux's goal is not to have a stable dev practice. ( see the whole spew about bin drivers.. :( )

Why do you guys think redhat has RHEL... to stabilize linux. go to any other distrib, and well.. things change often.

Fast change does not bode well with stability. Stability comes with time.

You want fast and cheap, go linux.

You want stable, you go commercial unix ( Solaris,AiX these days)

You want a good middle ground.. you go *BSD ;)

( yes, i'm biased, i've run extremely large bsd environments, but currently running a linux one.. and trust me, i miss my bsd )

AIX C compiler

[12357bd]

My first experience with AIX was auditing some large set of application C code. It was shocking, lots of uninitialized local vars, code assuming it to be 0, and it worked!
I suppouse someone at IBM decided to systematically clear stack var area at function entry... better that than to fix the broken code!.

Re:AIX C compiler

[Anonymous Coward]

They might have made a systemic decision that a) handing with random stuff from one process to another is a security hole

Different problem. The grandparent's talking about the compiler generating code that guarantees local variables are initialized to 0 on every function entry. The previous owner of that memory typically was some other function call within the same process, and there's clearly no security value in that case.

In fact, there's no security value if this is the first time the stack has grown to that size, either. Even if the memory came straight from the kernel, fixing it in the compiler's function entry code generation would mean that the process who is potentially eavesdropping on another is the one responsible for clearing the memory to prevent the eavesdropping. It's trivially defeated by using a different compiler on the malicious code.

To properly prevent this disclosure, the kernel must clear memory before handing it off to a different process (or at least to a different uid/gid). I believe that's done on Linux as well as AIX.

b) if customers' applications can be made more reliable and repeatable then that is a good thing and not something to be ashamed of!

There's a performance price for this choice. And it's typical of the attitude difference between Linux and commercial Unix - Linux developers would never accept sacrificing performance for the sake of some improperly written applications. (And then they'd go on to make some questionable performance choice that would break properly written applications, too. Like requiring fsync() on the parent directory for inodes to hit disk, ignoring that you only have to fsync() the inode itself on every other Unix system.)

Re:well

[Anonymous Coward]

Okay, I am both AIX and Red Hat certified so I kinda know what I'm talking about. When you are looking for a true enterprise-class UNIX, on which to base your true mission-critical applications, you've got to choose AIX over Linux every time. There's two major reasons for this, other than the "support from one vendor" argument:

- AIX, or more precisely, the Power5 (soon to be Power6) architecture has virtualization built-in the hardware, at the firmware level. Far more stable and efficient than VMWare, Xen or any other software-based solution.
- AIX is far better in supporting High Availability. The most important reason for this is that AIX has something called the ODM, or Object Data Manager. This is basically a list of all the hardware that's supposed to be in the system, and what the kernel needs to do with it. Including the possibility to detect, but not activate the hardware. If you are doing failover clusters, where certain pieces of hardware (e.g. storage) can only be accessed by one host, and one host only, you can tell the other host not to touch that hardware. And it will not touch it. Obviously this functionality is fully supported in IBMs enterprise HA product, HACMP. Contrast this to Linux, which scans all buses it can find, scans all adapters it can find, and then activates all the devices it can find, automatically. The only way to prevent your standby system to access the hardware is to "STONITH" (Shoot The Other Node In The Head, meaning forcibly take away mains power from the system). Crude.

Now combine all this with the single vendor argument, hardware/software/solution certification and validation and enterprise-class support worldwide, and you may understand why AIX is sometimes a better solution.

you forgot one

[ushering05401]

4. Bundle the fastest service response time w/said expensive solution.
5. Profit!

IBM made their rep w/me late one night in rural Vermont. I was troubleshooting my client's sole server (an ancient AIX rig) and shit started coming up wonky (hardware!?!?). This wasn't the sort of operation that had spare parts sitting around.

Worse yet, the client had all 14 of their locations (all running dumb terminals) running through this one server and their inventory and POS systems were going to be offline in the morning unless...

I still can't believe the response time for what had to be one of IBM's smallest, most outdated corporate clients. The IBM tech coordinated everything w/a third party on-site technician & we were up & running with shiny new parts in a matter of two hours (most of which was travel time)... Which gave me an hour or two to sleep before calling the company Pres in the morning to explain why they were going to have a big ole IBM bill in the mail.

You pay IBM for the absence of downtime, and it is worth every cent.

Regards.

Re:well

[LizardKing]

Linux not stable? Give me a break.

When my second to last employer switched OS from Tru64 to Linux, we saw a massive drop in stability. This wasn't a drop in stability or reliability of our applications, but of the OS and hardware. We had been an Alpha and Tru64 shop, and before that a Vax and VMS one. When the writing was on the wall after Compaq acquired DEC and HP then acquired Compaq, we switched to Linux on HP. This was their supposedly high-end machines, complete with huge RAID cabinets with dual redundant everything. From not needing to reboot the Alphas unless we wanted to reinstall the OS, we went to having to reboot the Linux boxes every couple of days. The RAID arrays would simply stop working, but more often than that Linux would go haywire and lock up with unkillable processes chewing up the CPU's. Despite a very expensive support contract, HP couldn't fix either issue, we just came to expect a visit from the engineer to replace the RAID controllers every so often and frequent reboots. As we were selling a logistics system to run warehouses 24/7, we were not happy and started to look at Solaris on Sun hardware. I left before the switch, but unless HP have managed to solve the Linux and RAID issues I expect that they have lost a customer by now.

Re:well

[walt-sjc]

I can see that certain combinations of hardware / software may not be entirely stable, but Linux in an enterprise environment can be VERY stable. I have multiple 8-way HP servers running Centos 4 with 50+ terabytes of storage (each) on an EMC SAN that haven't been rebooted since initial install about a year and a half ago. They systems get VERY heavy usage. Sorry you had problems with HP storage. We looked at HP storage and went with EMC for a number of reasons.

Now I HAVE had problems with a couple DL380 G4's and having them fall off the network occasionally (about once a month) due to some bizarre hardware / firmware issue, but only 2 machines out of about 100 have had that problem.

Re:obsolete?

[muirhejs]

I support 300 servers for a large financial institution. The cost of any one of them being down is up to $500,000/hour. AIX is our solution of choice. I love Linux and use it exclusively at home, but Linux simply isn't ready for this level of responsibility-- yet. We are starting to put some lesser-critical applications on Linux and we have it as an OS offering in our UNIX space, along with Solaris.

Some things that I'd like to see Linux achieve before it's really ready for prime time:
* Achieve a mature high-availability model. With the kind of uptime we require, I need a clustering solution that is very reliable and eliminates all single points of failure
* SAN support. SAN is still a relatively new (10 years or so?) technology. There are still quirks to work out and even Solaris and AIX occasionally have issues with them. It's a complicated technology. Add "Synchronous Data Replication" features and it gets more complicated.
* Drivers, Firmware, and Microcode. Because of the diverse hardware Linux runs on, I don't think enough attention has been paid here.

- John

Re:Ewwww

[Devir]

Linux is the roots to where I started my career back in '96. I was outwardly loyal to it for many years and at one time had a linux farm in my place. As long as i didnt try to recompile the Kernel or update the system I was good.

Then I discovered Free and OpenBSD. It was a miracle, I could create a slim custom kernel with almost no issues. I ran this for a few more years. Though I dreaded system update days. Tar configs, install new package, reconfigure test, pray. It'd be a weekend to just to update sendmail/postfix, spam filtering and the DNS servers. To update one thing you'd need to update ALL your libraries and underlying code base because everyone used latest/greatest.

Then I moved to a company that only used Solaris. This was a nice rock solid OS, though a little long in the tooth. While it was good, patching and updates were sort of difficult and lenghty process.

I dabbled in HP-UX as well. THis was a nice system, quick and easy to patch and update. The machine may have taken up a good portion of the computer room but it gave little problems.

Enter AIX. This is an OS that is a dream come. OS updates come in two flavors. The Maintenence/tech levels and the OS update level. If there is a failure in hardware or OS IBM can be onsite within 1-3 hours. They're always a phone call away, and the techs are well trained and knowledgable.

Smitty is their main tool for maintaining the OS. You can resize filesystems at the click of a button, tweak kernel parameters (sysctl in linux) and do a wide range of other things with ease. When I look at our few linux systems these days, I cringe to think i'd need to update them. While they run reliably, i dread patching and updating linux. I dread reconfiguring it. Ask me to patch and configure an AIX box and i'm all for it. It takes a few hours, vs the few days for linux.

With AIX you also have several options for OS imaging and installs. You can run an easy to configure NIM server, or simply run the mksysb command to backup the entire OS. Volume groups (filesystem groups) can be backed up just as easily. If the hardware fails, you can install the mksysb image onto another machine and be back up and runnning in 1-2 hours tops.

I know there are ways to make linux images and backup sets. It just needs 3rd party tools (last i knew) and is oftena pain.

I trust an airport running AIX. This isn't just million dollar planes, this is potentially my life at stake.

Wow! 138 Comments So Far...

[filesiteguy]

I find it actually funny that this is even really newsworthy. I'm sure the pointy-haired bosses at Quantas figured they'd save either time, money or staffing hours dealing with one vendor. Obviously they didn't want to go with MS in their server room, but they went from one *nix to another.

If you look a little further, you'll notice that the issue was with Financial operations. A few minutes with my good friend, google, turned up some tasty bits. For example here: http://www.fujitsu.com/global/casestudies/WWW2_cas estudy_Qantas.html

It says, "So when Qantas, Australia's largest airline, merged their international operation with a domestic airline and found themselves wrestling information among multiple data systems, something had to be done. The existing architecture was complex, slow, costly to operate and not very reliable. The response was IRIS, the Integrated Revenue Information Solution."

Guess what platform Fujitsu (the vendor) runs IRIS on...?