HP Embraces Linux for its Toughest Servers

Colmao writes "Investor's Business Daily wrote up an article interviewing Martin Fink, the head of HP's NonStop Unit. From the article'In a move that suggests Linux is finally ready for prime time, Hewlett-Packard is giving the free software a bigger role on some of its toughest servers.' NonStop servers are HP's most costly machines. They are designed to be always on, mission critical appliances. They are used to run some of the world's stock markets. Linux is making big moves in the datacenter and getting some much needed exposure."

Comments

Again?

[Anonymous Coward]

>in a move that suggests Linux is finally ready for prime time

Again? I think the last time was when it was let known that linux run several important systems in stock and other vital exchanges.

But of course...

[eno2001]

Linux is the OS most suited to big iron.

Re:But of course...

[cmstremi]

Site #32 is "www.kimberlykupps.com", #37 is "www.adults-online.com" and #50 is "www.wendywhoppers.com"?

I guess I'm not quite sure why I should be surprised that adult sites are up there on the list, but I am.

Perhaps netcraft factors boyancy into the ol' uptime equation?

correct link

[oringo]

The link given in the story is bad. There's a good story listed in yahoo news: http://news.yahoo.com/news?tmpl=story&u=/ibd/20050 720/bs_ibd_ibd/2005720tech01

Re:correct link

[FireAtWill]

Thanks for the correct link.

Having found the correct article, I'll be interested to see if this ever sees the light of day. I did spend a bit of time working on Tandem Non-Stop systems (since acquired by Compaq, and hence, by HP).

Working on Tandems was kind of like visiting another planet and seeing how an alien race might do operating systems. Hardware-wise, everything is redundant - from the CPUs, to the power supplies, to the system bus. Supposedly, a failure in any one component couldn't bring the system down. Everything was designed to survive a failure. Creating a fault tolerant system was supposed to be easy - except that all the software had to be designed to respond appropriately to a failure event.

The system could definitely use an upgrade using Linux. It was limited to eight character file names (no filename extensions) and you could use directories and subdirectories - only. There was no such thing as a subdirectory of a subdirectory. Just two levels.

The operating system was built around a messaging system that was fast enough for ATM transactions, but was useless for batch type data crunching operations (unfortunate, since somebody at this shop had selected it to do exactly that task.)

If they're thinking about this on the same lines as the original vision of the NonStop operating system, the version of Linux that comes of this will be completely non-applicable to any hardware but HP's. Everything will need to be rewritten/tweaked. Starting with Linux will help their other deficiencies, but will result in a really weird Linux derivative.

But in the end, IMHO, the whole concept is flawed because, while interesting, it only addressed component failure and not disasters such as fires or tornados, etc. At the same time I was working on this system, I read a story about a mass transit system in Denmark automatically (and successfully) failing over to another system in another city as a result of a fire. That seemed far more robust. And those systems were running OpenVMS.

Re:correct link

[iGN97]

I currently work with the HP NonStop on the server side doing financial transaction processing.

While some of your points are indeed valid, your post loses some value because it's incomplete. True, Guardian, the underlying operating system, has "odd" limitations, like eight-character filenames, etc, but most applications we write these days uses OSS, which is a POSIX-layer on top of Guardian, allowing for "normal" UNIX-style filenames. True enough, there's always something that's different enough to make straight ports of UNIX software difficult, but the work involved is usually minimal.

The key strength (IMHO) of the system, is the "pathway system", which is a transaction based, load balancing message passing system which allows you to scale an application close to infinately, across physical machines and sites. It's simple to monitor, and it allows you to see which server processes need more instances easily. It's also very easy to setup more server instances, and your application code doesn't necessarily need to take extra steps to be instantaneously scalable. It also ties into the SQL-based databases which run on the system, so errors can be effectively backed out of.

Regarding disaster-recovery, I would hardly call HP NonStop "flawed". We have a separate physical site in case of fires, bombing, etc, and although there's no "automatical" failover setup in our facilities, failover from one physical site to another is an important piece of the NonStop design, and we rely on it. It's also very convenient to have another site to run an application when doing major upgrades, etc.

Next generation NonStop-machines will also be Itanium based, which, IIRC, will allow application programmers to use Intel C/C++ which is great at optimizing and very good at conforming to standards. That being said, the current line of development tools (ETK) allow you to write C/C++ with embedded SQL on the PC, hosted in Visual Studio, cross compiling with built in deployment-features using FTP. I think it's a fairly nice environment to work with considering the age of the hardware this is running on.

If you're in the market for a platform to do massive transaction based processing, you'll do yourself a favor by considering HP NonStop.

Always on servers?

[jrockway]

What will they think of next!? Personally, I shut all of my servers down at 5PM so that people working late are inconvenienced and all incoming mail delivery fails! Take that, SPAM!

Oh wait.

Re:Is that "NonStop", as in Tandem?

[bleak sky]

Are we talking about the remnants of Tandem being moved to Linux?

Yes, it's NonStop as in Tandem. But it's not so much about moving it to Linux, but enabling Linux to run on it (adding NonStop features to the Linux kernel), and allowing more open source projects to run on NonStop.

Re:hp double faced?

[drspliff]

Eh?

HPs Desktop business is dealing commodity hardware for 'mom and pop' kinda people who need to check their e-mail, browse the web and share videos with family etc.

I can totally see why Linux is unsupported on their desktop systems, it's a pure business decision due to the relatively tiny number of Linux users buying their systems.

On the other hand their server business is the exact opposite due to the increasing market share Linux is getting in the data centre. Linux has already proved it's self on their entry and mid-range servers for a number of years now and their finally giving it the break into mission-critical data centres that it deserves.

Looking at the parents comment they have never dealt with HP servers running Linux, or indeed HP servers running anything. The platform support package (PSP) is great, it includes industrial strength drivers for their RAID cards, power management interfaces and even utilities to toggle the maintainance LED.

All in all HP could be called double faced, but the amount of development work required to make/certify drivers for all the desktop hardware they make just isn't worth it just to persuade the few Linux users that haven't heard the HP Desktop horror stories to buy their systems.

Re:Is that "NonStop", as in Tandem?

[jcr]

I would be very apprehensive about trying to graft the NonStop technology into the Linx kernel. Running Linux as a process under Guardian, just like IBM runs Linux under VM, makes a lot more sense.

-jcr

Re:hp double faced?

[heeeraldo]

you bought a car?!

I built my own from parts I ordered from newegg.com! Saved me a fortune, although I still can't find a distro of RedHat optimized for a small-block Chevy...

HP Needs Linux to Survive

[reporter]

HP Unix was distinctly inferior to IBM's AIX, and moving to Linux is a way for HP to (1) jettison its own inferior product and (2) avoid the R&D costs of developing a superior operating system (OS). As for #2, there are 2 aspects: (1) feature development and (2) reliability. The vast army of open-source developers have made Linux a feature-rich OS that rivals (and likely now exceeds) IBM's AIX.

Further, IBM has spent enormous sums of money to ensure that Linux is reliable. IBM will soon discover that this aspect of Linux is the Achille's heel of open source. By using Linux, HP essentially gets a free ride from IBM and need not spend the money to ensure that Linux is reliable. IBM has already done the work.

I can already hear the grinding of the reduction-in-force axe at the OS department of HP.

Re:HP Needs Linux to Survive

[Compuser]

Don't worry about IBM. Having core developers in-house
boosts their services part. If HP cuts their devs
and goes with Linux without R&D part in place then
their efforts to develop their service business
(something they dearly want) will hit the wall sooner
rather than later.

Re:HP Needs Linux to Survive

[nihilogos]

Further, IBM has spent enormous sums of money to ensure that Linux is reliable.

They haven't even spent a fraction of the amount of money that they would have in developing their own operating system from scratch.

IBM will soon discover that this aspect of Linux is the Achille's heel of open source.

I'm sure they were already aware that contributing to a GPL project means other people can use your code.

By using Linux, HP essentially gets a free ride from IBM and need not spend the money to ensure that Linux is reliable. IBM has already done the work.

I doubt IBM spent any time worrying about how to support the sort of redundancy that goes into the NonStop servers. HP would have had to contribute a lot of that themselves, and guess what? IBM gets access to all that code.

Re:HP Needs Linux to Survive

[WindBourne]

First, I have worked for both HP and IBM.

Even though I am a die-hard Linux coder, I would say that HP-UX is NOT inferior to AIX. I have worked on both, and I would say they are the same. What really matters is the support that the company offers. Can you count on the company to provide patches quickly? Is the hardware solid. Once again, I think that both companies make good equipment (but on the lower-end, I would only buy from one of them).

As to the achilles heel of open source, IBM is much brighter than you about this. Several things:

1. Other than MS, ALL actively supported OS are money losing operations. Even for Apple, they lose money. Once active support stops (i.e. about 6 developers), then the company makes money (HP has made millions off HP-3000 once they stopped active support). The OS exists to sell hardware.
2. HP has been contributing to the OSS world for some time. They may get a bit of a free ride, but they will also have to contribute more or will be unable to say that they offer the best support. IOW, IBM and the other companies will be having "free" rides.
3. It is EXPENSIVE to create an OS and then run in on only one platform. That is how MVS/CICS/etc is on the mainframe. Or how OS-400 on the as-400. Or AIX on power-pc systems. Linux is on all of their systems (and their competitiors). Soon, Linux will be #1 for IBM due to ability to scale across the hardware. If an Intel does not cut it, then Power-PC it. If needing more, that AS-400, followed by the big boy.

Yes, HP will get some free software, but IBM is getting 10s of millions of support from everybody else.

As to the axe, well it will fall in all the major tech. companies. They are all pulling a fast one. In the past, they would lay-off in the states or in EU, and openly hire in India/China at the same time. Now, they are going through fast up and downs. Well if you watch carefully, the up is hiring in India/China, and then 6-9 months later, they announce a slow-down and lay off. You are simply looking at the shifting of ALL tech companies to overseas. IBM and Sun will soon announce another round of layoffs. While the American economy is still lousy, so you will not notice.

Found TFA!

[Russellkhan]

The original article seems to be inaccessible on investors.com. I found the same article on Yahoo news.

Nice story, bad title.

[ron_ivi]

The story's titled "HP Propelling Linux Into Truly 'Big' Time"

Considering

• Linux is the leading OS in the Top-500 supercomputers, and
• Linux runs large clusters such as Google, and
• Linux runs a bunch of stuff for Schwab, ETrade, etc -
• and this other computer company that's a bit bigger than HP called IBM already noticed Linux

I think this article is badly misnamed.

The article should have been titled

"Linux Propelling HP into Truly 'Big' Time".

Re:hp double faced?

[jschottm]

I own a hp laptop and i cant get some of it's features to work under Linux. ...
They put on their most expensive hardware an OS that they don't support.

What to make of this?

HP is a massive company with ~150,000 workers (minus those cuts that are about to happen). The team that does the very high end systems discussed in this article have very little to do with the team that designed your laptop, other than getting a paycheck from the same company. They have far different interests and customer needs than the laptop people do. Linux has very limited penetration and market share on laptops but a large and increasing share of the server market.

What is Non-Stop

[minniger]

These are systems that are really pretty cool. And really freaking expensive.

They have nothing to do with HP-UX or Unix of any kind. They are Tadem machines (feel free to look that up).

These are rather slow but super reliable machines with a bizzare OS that has had features for decades that mainstream os's still don't have. Take the current clustering and grid tech and meld it all together and you get something like the tandem. The company I work for came out of the tandem space. The typical intro to the machines for new hires is to note that you can smash one with a sledge hammer and you won't lose any transactions.

Who uses these things? Banks, Banks, Banks, Airlines, Governement, Dell, etc...

They (HP) have been working on a unixy layer to run on top of the tandem os for a number of years now. Apparently this hasn't been going too well. Sounds like Linux might help them do something similar to IBM and the VMs on the mainframe.

Re:NotStop != non-stop

[frodo from middle ea]

Originally The System was called Tandem, which ran an OS (or rather kernel )called "NonStop Kernel", they even have a database called NonStopSQL.

Later in S-series servers, the System got renamed to NonStop, the kernel got renamed to Guardian.

About loosing CPU,Memory/System Bus you are not even scratching the surface....With 2 systems operating in Tandem (hence the name),, you can even loose an entire system, and the other one takes over, and these 2 systems can have 2 geographically remote systems in tandem, giving you complete fault tolerence.

These systems talk to eachother over a proprietory network stack (defi. not TCP/IP) and do health monitoring. And uptime is measured in years, not months...