Today we host an interview with Christophe de Dinechin, Software Architect in HP-UX (Software business unit, Infrastructure Solutions). Most of you already know HP-UX, the leading “traditional” UNIX today feature-wise (second only to Solaris in Unix market-share, mostly competing with AIX). With Christophe we discuss HP-UX’s competition, the other… 5 OSes HP supports with its various products, the Itanium platform and more.
1. What is HP-UX’s main market? Where does the OS is mostly used?
Christophe de Dinechin: The short answer is: “enterprise computing”. The long answer can be found on HP’s web site.
2. What are HP-UX’s advantages and disantvantages over AIX and Solaris, purely technically-speaking?
Christophe de Dinechin: HP-UX has been ranked #1 overall by the DH Brown 2002 Unix review. So it’s a well-balanced Unix.
But in my opinion, the major differentiator is Itanium. An OS like HP-UX, AIX or Solaris cannot be considered in isolation of the platform it runs on. And Itanium is taking the top spot in performance regularly, from SPEC to TPC, yet you can buy Itanium workstations for less than $3500. So customers are starting to believe HP’s “high performance at low cost” statements.
3. What do you see as the biggest competitor today on HP-UX’s business? Windows or Linux and why?
Christophe de Dinechin: Both are very serious competitors. If you are caught between a lion and a leopard, you don’t start asking “Who’s the most dangerous”, you think about ways to fight both or have them fight each other.
Linux competes more in the technical computing area. Itanium supercomputers run Linux. The development environment and ecosystem of technical solutions makes the difference.
Windows competes in the traditional “enterprise” markets. They got very good TPC benchmarks lately.
4. What features from Tru64-UNIX have been or will be incorporated to HP-UX? What is to happen to Tru64, will they be new big releases or just patches from now on?
Christophe de Dinechin: HP released a roadmap for a transition from Tru64 to HP-UX, and also published a FAQ which answers these questions. According to this FAQ, HP is committed to develop new versions until at least 2004, including support for up to 64 processors, and transition tools to HP-UX. There will be at least a couple of point releases, not just patches. Naturally, during that time, HP-UX will start integrating some of the strengths of Compaq’s offering.
5. HP supports a whole range of OSes, from VMS to Tru64 and Linux, Windows etc. Do you believe that each OS is built and is scalable to a specific market, or do you think that all OSes can be engineered to do (and do well) just about anything?
Christophe de Dinechin: Linux and other GNU operating systems (BSD, Hurd-based, etc) share a remarkable flexibility due to their open-source nature. They can run on anything from watches or toasters to supercomputers. So an OS can be engineered to do just about anything.
The downside is that the GNUs tend to not do anything particularly well, with a few exceptions (BSD’s security is an example). Linux doesn’t scale exactly as well as HP-UX, and it’s user interface is not yet as consistent and newbie-friendly as MacOS X.
Commercial offerings on the other hand have a “sense of purpose”, a strategic direction. They are built for a specific market because they need a market, they need to make money. Linux doesn’t need to make money.
The boundary is quite fuzzy, however. Microsoft is ready to lose a lot of money to conquer a new strategic position, see their current work on embedded systems, PDAs and cell phones. Similarly, HP and Intel invested massively in Itanium. In the brave GNU world, Linux is getting strategic directions from the people who want to make money with it, like HP, RedHat, IBM.
6. Recently HP canned the Gnome 2 porting project to HP-UX. Why was that? Do you feel that CDE fits the bill as a modern desktop environment today? Are there plans to replace CDE with something else?
Christophe de Dinechin: I have no idea why this decision was taken, but I guess it has something to do with the fact that the workstation market isn’t that hot right now. I personally don’t like CDE much. The Unix Hater’s handbook rightfully calls this kind of software: “complex non-solutions to simple non-problems”. But then, I’ve not been thrilled by Gnome or KDE either. When I run a GNU, I tend to stick to good ol’ WindowMaker, just because I don’t feel it’s worth changing all my configuration files.
But using Linux as a desktop happens less and less for me, since all my home machines run MacOS X these days. Now, _that_ is a GUI I’d like to see on HP-UX. Steve, Avie, Bertrand, don’t you want to port to HP-UX once more? Pleaaaase?
Ah, wait, I take that back. I still have a Linux machine at home. My CD player runs Linux. It’s an HP DE200C. It would have been a darn cool machine if it weren’t for the price (close to $1000 for a CD player? No thanks!)
7. Today HP is pushing HP-UX mostly on Itanium processors and not on HP PA-RISCs. Do you feel that “sharing” the platform with Windows and Linux can be as successful as running on your own platform? Also, are all OS features available in the Itanium architecture as they are for the PA-RISC one?
Christophe de Dinechin: Competition is good. And HP has placed bets on all three runners anyways. HP is always very pragmatic So, as long as we can maintain a strong value proposition with HP-UX, customers will keep buying it. On the other hand, there is no law of thermodynamics stating that HP-UX business must always increase. Hmmm, I guess I just made a comparison between HP-UX business and entropy there :-/
In terms of features, the gap is closing rapidly. There are still a few missing features, like virtual partitioning (vPars). But we are supposed to reach at least parity soon, and we are working on it.
8. How do you feel of the open source movement? Do you think that releasing the source of HP-UX or any other of the HP-developed OSes is compatible with your strategy or not?
Christophe de Dinechin: I have a significant investment in open source myself. But free software or open-source software doesn’t always make sense. Again, HP tends to be very practical.
To take your example, I’m not sure what open-sourcing HP-UX would bring to customers. One of the key values of HP-UX is reliability, as in “if it crashes, if it doesn’t run well, if there is a missing feature, we take the blame and we help you”. It’s more difficult with the GNUs. Also, HP-UX contains a lot of proprietary code, so I’m pretty sure we couldn’t open-source it if we wanted to.
On the other hand, if you want to know the details of the hardware, you can look at HP’s contributions to the Linux kernel. David Mosberger and Stephane Eranian ported Linux to Itanium. In that sense, Linux on Itanium is an “HP-developed OS”. And then, David and Stephane wrote a book to explain how they did it, and HP published it. If you don’t have this book, go buy it. Now! It’s really good for anybody interested in OS design. And I’m not just saying that to help my friends
9. How’s the port of VMS to Itanium is progressing? When will it be ready for general consumption? Will you continue offering the OS under the “hobbyists” license?
Christophe de Dinechin: Frankly, I can’t tell you. Both because I don’t know much and because I would have to kill you if I told you.
10. How do you see OSes getting evolved with time? What in your opinion, OSes will be able to do in 10-20 years from now that today can’t, technically-speaking?
Christophe de Dinechin: Three things:
– The OS itself will probably fade into the background where it belongs. You don’t care much about the OS of a Palm Pilot or a network appliance or an ATM, and you shouldn’t. The OS would probably have disappeared from the public consciousness five years ago, weren’t it for Microsoft’s insistence on making it its main source of profit (so people have to _know_ what an OS is, otherwise they wouldn’t want to pay for one). Illustrating this trend: Java, OpenDoc (no OS, no applications, only documents), the Taos operating system (I hope that http://www.tao-group.com still explains somewhere what this architecture-independent OS was), the Linux compatibility layer in most Unix variants including HP-UX, Virtual PC for the Mac, etc. In twenty years, I hope you will be able to run applications without having to think about the OS.
– In terms of user interfaces, I believe that vocabulary-based user interfaces will emerge. With a mouse or a pen, you can only select about 10 commands at a time, and this gets more constraining as devices get smaller. With words, you can select about 10,000. It doesn’t have to be words, however. A voice-mail system where you tap “1221” to get your last voice mails without even thinking about it is a simplistic form of what I call a vocabulary-based user interface. At the other hand of the spectrum, what about thought-driven user interfaces? It might come sooner than you think to a PDA near you. By the way, thinking about how to program these things ultimately is one of the things that drove me to develop Mozart.
– One trend has almost never been broken: computers take more and more time to boot. In 20 years, expect that the average PC will take anywhere between 2 hours and 10 days to boot. Don’t laugh, we would not have believed in 2 minutes boot times back when we had instant-on Apple IIs. But then, humans take a few years to boot themselves, there’s a lot of hardware diagnostic tests to run (Test emergency alarm system; Test input processor; Test output ejector; Test low-power mode; Repeat).
He keeps referring to *ALL* open source *nix clones as “GNUs”
Sorry dude, only Linux and HURD should be called that name.
The BSDs are an entirely different beast all together.
(And GNU zealots, please don’t start with the “*BSD wouldn’t be anywhere without GCC” bull-crap)
The Apples were instant-on? Oh my god! BeOS is too slow now!
If BeOS would have all the feature included in OS X 10.2.6, it would be much slower. It’s easy to boot fast when an OS doesn’t do much.
If you have an ACPI system, just put your system to sleep (suspend to RAM) and you get a near instant start-up, plus no need to close down all your apps. I’m hoping once Linux 2.6 comes out someone will develop a nice GNOME GUI for controllong stand-by settings, etc.
A very intelligent man this Christophe de Dinechin =)
A whole lot of the links are broken.
Great interview, by the way. I use HP-UX and GNU/Linux all day. Nice to see that someone at HP cares about their product. Too bad GNOME is not going to be supported – CDE is ridiculously dated.
>A whole lot of the links are broken.
All links are working for me.
Machines keep getting faster, boot times haven’t really changed much since windows 3.1. Nobody will buy a product that takes a few hours to boot.
This guy sounds dumb with a capital D.
http://www.openvms.org/stories.php?story=03/06/30/5087031
If BeOS would have all the feature included in OS X 10.2.6, it would be much slower. It’s easy to boot fast when an OS doesn’t do much.
I think you meant to say, “It’s easy to boot fast when an OS isn’t saddled with decades of cruft”.
HTH!
The guy likes WindowMaker. Sweeeet. He obviously is smarter then the masses running the bloated heaps of shit know as Gnome and KDE.
Plus, any company that has liquid cooled CPUs in their boxes is a-ok with me.
Good interview. But I disagree in some parts in special the “unfortunally paragraph”: “The downside is that the GNUs tend to not do anything particularly well, with a few exceptions (BSD’s security is an example). Linux doesn’t scale exactly as well as HP-UX, and it’s user interface is not yet as consistent and newbie-friendly as MacOS X. ”
If GNU not do anything well, why HP/UX packaged Open Source applications? HP/UX have a better Web server than Apache (I know that is not GNU but is Open Source)? This friend doesn’t use GCC?
In the same phrase, please remember that MacOS X is based in FreeBSD.
In the scalability side, as he knows, NEC has probed Linux scalability in one 32 Itanium 2 server and Intel report 600.000 tpmC in one 32-way Itanium 2 Linux server. Is not 600.000 tpmC @ 32 procesors very amazing compared to 750.000 tpmC @ 64 proc?
Regards,
Bryam
If GNU not do anything well, why HP/UX packaged Open Source applications? HP/UX have a better Web server than Apache (I know that is not GNU but is Open Source)? This friend doesn’t use GCC?
He uses ‘GNUs’ to refer to the GNU/Linux operating systems (and also BSD’s, etc…) and not to GNU applications like gcc and what have you. Read and understand.
“Do you see HP-UX eventually being phased out in favour of Linux?”
HP are using Itanium to migrate their existing PA-RISC customers over, but it’s a double-edged sword – attracting new customers to HP-UX on Itanium is going to be very tough when there’s BSD, various Linuxes and even Windows that can run on exactly the same Itanium hardware as HP-UX does.
In the long run, you can see Itanium Linux dominating the 64-bit server platforms, taking market share from every one of its rivals, even Solaris. Yes, there’s AMD efforts which provide 32/64-bit backwards compatibility (Intel are catching up with that towards the end of the year), but in 5 years time, I suspect the Itanium family will be the #1 64-bit server chip out there. Linux will have 50%+ of that 64-bit server market and the rest doled out between Solaris and Windows primarily…
Yes, there’s AMD efforts which provide 32/64-bit backwards compatibility (Intel are catching up with that towards the end of the year), but in 5 years time, I suspect the Itanium family will be the #1 64-bit server chip out there. Linux will have 50%+ of that 64-bit server market and the rest doled out between Solaris and Windows primarily…
I guess that explains why you’re not an analytic. Linux might have some small market share as today but geeez, 50%? More like 5%. It’s not even close to offering what Solaris offers…
See if things are good, people who know that use them. Period.
He’s talking about servers, not about desktops…
Mr. Bryam wrote: In the scalability side, as he knows, NEC has probed Linux scalability in one 32 Itanium 2 server and Intel report 600.000 tpmC in one 32-way Itanium 2 Linux server. Is not 600.000 tpmC @ 32 procesors very amazing compared to 750.000 tpmC @ 64 proc?
Well, no. Clustered vs. non-clustered results on the TPC-C test are not comparable because TPC-C is so darn easy to partition. For example, a database running TPC-C on a NUMA cluster of 32 two-way systems would kick in the teeth of a real 64 processor box, ceteris paribus. That stack of dual processor systems is certainly cheaper, too. But the 64-way system would crush the cluster on anything that couldn’t be partitioned the way TPC-C can.
Yours truly,
Jeffrey Boulier
PS Any chance that the PRE tag could be added to the “allowed HTML” list? Can’t think of any quick way to explain the concepts without pretty ASCII pictures. :0)
In response to someones preemptitive strike against GNU, I would just like to extend my gratitude to all free and open source developers, terms which are used interchangably, as I am very grateful for the tools they have made available to me such as TCP/IP, GNOME, GCC, KDE, and other things under a myriad of free licenses, both copyleft and non-copyleft, respectively. I love my freedom and see no reason to attack anyone who has helped to ensure those freedoms, such as the Linux, GNU, and BSD projects.
Let’s all keep in mind that this guy is an HP-UX Engineer.
I think it would be a little much to expect him to gush about how great Linux is and how HP-UX is doomed with a capital D.
I won’t buy anything from HP. My mom bought a monitor from them for $500 and it died on her two years later. It was worse quality than her eMachine. I’m not buying a workstation from some comapny who can’t provide quality across the board. That is the point of a brand. If I buy something entry level from Sun or SGI I can be sure it will be top of the line (even if it is Intel or Linux).
OS X is based on Next interface. It uses the Mach kernel. It uses network components as part of FreeBSD. The insides are based on Darwin and is GPL.
>The insides are based on Darwin and is GPL.
No, it isn’t. Darwin is released under Apple’s APSL.
Mr Jeffrey Boulier wrote: “Well, no. Clustered vs. non-clustered results on the TPC-C test are not comparable because TPC-C is so darn easy to partition. For example, a database running TPC-C on a NUMA cluster of 32 two-way systems would kick in the teeth of a real 64 processor box, ceteris paribus. That stack of dual processor systems is certainly cheaper, too. But the 64-way system would crush the cluster on anything that couldn’t be partitioned the way TPC-C can. ”
It was one non-clustered result: http://news.com.com/2100-1010_3-1013764.html?tag=fd_top
Regards,
Bryam
I looked at the article. Very interesting. Unless I missed something though, I’m not sure we can say for sure whether or not the database was clustered — running flat out oracle or OPS the way the Sequent boxes did. It seemed very coy with all details of the test.
Does Linux support NUMA?
Anyway, still a ways to go to catch IBM. 32-way power 4 @1700Mhz w/ DB2 scored 764,000. November availability. Anyone want to give me $6,349,223 so I can buy one and run my own tests?
Eh, Sun has me sufficiently brainwashed that I ignore TPC-C. I’d still take the six mill though.
Incidently, Oracle on Linux/x86 has really been improved. Oracle 8.1.7.4 was a step backwards from 8.1.7.3, which wasn’t itself so hot. But 9i is really sexy…
Yours truly,
Jeffrey Boulier
There’s a fair bit of Numa support in the development kernel from what i remember, go to
http://kernelnewbies.org/status/latest.html
for everything new that has being added into 2.5 since it started.
<EM> He keeps referring to *ALL* open source *nix clones as “GNUs”
Sorry dude, only Linux and HURD should be called that name.
The BSDs are an entirely different beast all together.
(And GNU zealots, please don’t start with the “*BSD wouldn’t be anywhere without GCC” bull-crap)</EM>
I respectfully disagree, but I admit that my choice of nomenclature is largely a matter of personal preference.
Stallman’s point about saying “GNU/Linux” is that the GNU movement was started to create a free Unix clone, and that this included both the applications and the kernel. Today, on any free OS, including BSD, you find mostly the same (user-space) stuff, and a lot of it can be attributed to GNU origin. Apple ships tons of GNU tools with their BSD-based MacOSX. Only the kernel really differs. The rest is (according to gnu.org) about 30% GNU, and 70% the rest, including the kernel. In all cases, on a typical system, GNU in general represents the most important single contribution (even more so if ranked by usefulness rather than by line of codes, IMHO).
In that respect, I see really no difference between BSD and Linux. This is the reason I disagree with your comment. BSD may have had self-standing implementations with zero GNU contribution. But today, any BSD system needs GNU stuff to be really usable. I probably wouldn’t use OSX much without bash, gmake, gnu tar, gzip, gcc or emacs!
By the way, I don’t really like calling Linux systems “GNU/Linux”, simply because it’s long in the mouth. And it sounds petty, if you ask me. On the other hand, I think that the GNU project deserves some recognition. Calling all “freedom-compatible” systems with a large GNU genom “the GNUs” seems like a good way to remind people of the GNU contribution. And it’s short, and it sounds good.
But, again, that’s just me.
Didn’t anybody ask :
1. But wasn’t the Alpha a better chip than PA-RISC will ever be (moot now they’re moving to Itanium but still . . .) ?
2. Isn’t Tru64 still a better Unix than HP-UX will ever be (probably had more customers too before Compaq/HP started to provaricate over the future OS roadmap) ?
๐
Jeffrey Boulier wrote: “Does Linux support NUMA? ”
A pair of link:
http://lse.sourceforge.net/numa/
and
http://home.arcor.de/efocht/sched/
Regards,
bryam
Well, I should point out that out of the small list of GNU software you provided only gcc, emacs, and bash do not have identical BSD versions. The entire BSD userspace, ps, top, tar, etc are all BSD licensed programs with no help from GNU.
If you need emacs, gcc and bash to make a system usable, then nearly no system including HP-UX can be considered usable without GNU. Sure, gcc is not needed, but emacs and bash have no non GNU versions. There are certainly alternatives, but the BSD systems come by default only with those alternatives anyways. The only thing the BSD systems rely on GNU for is gcc. It’s very unfair to call them GNU systems as certainly gcc does not constitute 30% of the system.
Besides, the real reason people get upset when BSDs are linked with GNU is the differing philosophies. The GNU people are very much anti-corporate or anything non-free in the FSF sense. Many of the BSD developers are not out there to topple any corporation, but rather just to provide a good system. Also, many BSD users do not consider GNU software to be ‘Free’ in comparison with BSD based software.
I suspect that’s where most of the tension is coming from.
Thank You!!!
I couldn’t have said it better myself.
I’m also curious what the 30% is in reference to…lines of code, bytes of code, bytes of binaries, number of binaries, or some other metric entirely?
I have to admit, I do like bash…but emacs…YUCKY!!!…LOL
And gcc is nice for the very verbose error messages and its ubiquity if nothing else (not quite as nice as IBMs Jikes Java compiler though, damn I love that piece of software)
If “decades of cruft” means everyday features, I rather have the cruft and leave BeOS behind, thank you very much. Frankly, if BeOS had a vector graphics system – would it be near as fast as OS X? Well, it might be faster, but not by a whole lot.
You may have a bad experience with HP, but I don’t think it is fair. Many monitors though matter the brand, both LCD and CRT in very very different ways die out faster than counterparts even from the same model family! If everyone could be guarenteed 100% uptime for monitors, that would be an engineering feat not comparable to anything done before.
Besides, if you refuse to buy a server from the same company that makes your spoiled monitor, forgive me for saying that you are plain stupid. There are many different autonomous departments of HP, the quality of servers aren’t affected in any way by the poor quality of HP monitors – they are from two very very different set of engineers, manufactured in different locations, etc.
Well, it is your consumer choice, but frankly, you are limiting yourself from one very competitive company. Your loss.
Tru64 is one of the worst Unixes in stability and features. I don’t remember who did the ranking, but HP-UX was the best and the lowest ranked outside of SCO and the Linux world.
I can’t remember who, but someone said tru64 sucks so it must be true…
Interesting analysis of a solid O/S that continues to be a leader in stability and features.
TruCluster continues to be the best clustered UNIX offering, way better than MC Service Guard.
I have my doubts as to whether or not HP can deliver on it’s promise of incorporating tru64 advanced features into HP-UX.
Markets dictate which products live and which ones are retired, but I’m not alone in the belief that tru64/Alpha is/was far superior to HP-UX/PA-RISC.
As an alternative for bash, there is the standard sh, which is not as enhanced as bash, but still a program like bash.
For the rest, I find it stupid that, while BSD has yacc and make, many programs need bison and gmake to compile.
And then, in some ways I do not like GNU. For example, if you download findutils it will not compile. This is a known bug, but it has not been fixed in five years! There has not even been an update in five years!
The GPL can be restrictive, too. And too long and not-understandable. For example: if I write a package under the GPL, I need to also place the installer under the GPL. But what if I want to use that installer with another product, which is not GPL? That would not be allowed if anyone except me would make a contribution to the installer.
Did anyone notice that he admitted that he prefered to use Apple OS X as a desktop at home? Very interesting, though I am not personally an Apple fanatic, and my PC and printer are coincidentlly from HP. ๐
I also was a bit annoyed by his calling BSD “one of the GNUs”. Even RMS wouldn’t call BSD “GNU/BSD”; although it assimilated some GNU tools much of it’s userland and all of it’s kernel is BSD. (If it weren’t for gcc being better, for example, it might still be using a non-GNU open source compiler early non-encumbered *BSD came with; and *BSD’s libcs are not from GNU either…) I guess such terms as “the GNUs” make sense if you’re trying to distinguish a propritary Unix from the free competition to laypeople, but he should have known better than to make such gross oversimplifactions on a OS enthusiast site. ๐
Raj wrote: [i]Isn’t Tru64 still a better Unix than HP-UX will ever be (probably had more customers too before Compaq/HP started to provaricate over the future OS roadmap) ? [i]
I don’t know about better, but it certainly didn’t have as many customers. The story goes that while the VMS move from VAX to Alpha showed how well a transition of architectures can be managed, DEC’s Unix transition was handled very badly. As a result, Digital Unix slipped into the middle tier of Unix vendors and never fought its way back to the top.
Oh, and thanks for the NUMA links, guys!
Yours truly,
Jeffrey Boulier
A bit off topic but the guy who complained about the HP monitor lasting 2 years……….
I have had 2 Acer monitors-one lasted 4 years the other lasted 6 months.
Its all a matter of how they were built and mostly good luck.
But to put down every product HP makes because of one faulty product you bought is insane. For all you know the monitor could of been built by some other company and had a HP badge stuck on it.
Companies dont always make all thier products themselves-esp in the world of IT
>DEC’s Unix transition was handled very badly. As a result,
>Digital Unix slipped into the middle tier of Unix vendors
>and never fought its way back to the top.
We write software for the Alpha/Tru64 and its been fairly painless – we’ve even managed to help Compaq/HP sell decent spec servers (multiprocessor ES40/ES45’s and better) on the basis of how well our compute codes run on it. For our sector (molecular modelling, informatics) raw power is paramount and this platform certainly delivers.
While the Itanium is good (Itanium 2 is even better) at this early stage its still only a fraction of the Alphas speed. Intels bug-ridden compilers don’t help much either . . .
At a rough guess I’d say HP-UX was placed just behind Tru64 and Irix in the mid-tier Unix world (behind Solaris and AIX) in terms of sales.
Lets face it – no one ever sang the praises of HP-UX and PA-RISC – theres either a massive marketing problem or the arch just isn’t very good.
Still the Alpha & Tru64 are now heading to the great processor / os graveyard in the sky . . .