Oracle loses appeal in HP row over Itanium Oracle's last-ditch effort to wriggle out of a judgment requiring it to continue support for HP's Itanium-based servers has failed, leaving only the issue of damages to be resolved. In August 2012, a San Jose, California court ruled that Oracle had violated the terms of its contract with HP when it announced that it would no …
".....So, now you're slagging off the ability to disable some cores for higher performance. Is that because Itanium can't do it?...." What I'm slagging off is IBM's deceitful way of pretending their benchmarks reflect real world performance. Itanium can adjust core frequency and switch cores in and off, but what hp quote is the minimum frequency you will see the chip perform at. They also don't try and hide the fact that switching on and off cores will incur licensing costs. IBM always go on about the maximum frequency they can get from a core, hiding the fact that is with half the cores disabled, and will deny the licensing hit right up until you show them a letter from Oracle stating the facts! I've been there, you obviously have not.
"...... If you look on TPC-C and SpecInt...." Oh dear, you're not a regular here, are you? We already have two resident IBM benchmark drones, Jesper and Alli, and they have been getting debunked for years. I suggest you go read some of their posts and the replies before you look really stupid.
"What I'm slagging off is IBM's deceitful way of pretending their benchmarks reflect real world performance. Itanium can adjust core frequency and switch cores in and off, but what hp quote is the minimum frequency you will see the chip perform at. They also don't try and hide the fact that switching on and off cores will incur licensing costs. IBM always go on about the maximum frequency they can get from a core, hiding the fact that is with half the cores disabled, and will deny the licensing hit right up until you show them a letter from Oracle stating the facts! I've been there, you obviously have not."
Not sure who your IBm sales rep is, but I've never felt deceived by anything ours has said. I've never had anything like you're describing above. Maybe he just doesn't like you? Can't imagine why!! How has IBM ever hidden that switching cores on and off will incure license costs? IBM has never spoken to use about license costs. They tell us to read our Oracle agreement!! All the literature I've seen also quite happily has both the TurboCore and MaxCore frequencies listed independantly, so where's the hiding?
No benchmark ever reflects real world performance. Not Oracle Sparc, not IBM and not HP. The current trend is to shove SSD everywhere to increase the I/O performance and therefore do better at TPC-C. There were other methods in the past. Anyone who believes benchmarks as real world also believes in the tooth fairy and car MPG figures!! Maybe your rampant hatred of anything non-HP is shining through, so they just decide to wind you up!!
The Register is credited with the origin of the Itanium/Titanic joke (which is strange I as I was convinced it was part of DEC's desperate attempt to keep Alpha alive?), way back in the day when they were still storing their editorial head up McNealy's rectum. Indeed, during what must have been the most drawn out and unsinking sinking in history, El Reg has stuck by their petite blague, even during the editorial changes that led to their shift of rectum to Palmisano's. Unfortunately for Oracle, Larry doesn't seem to have realized that he needed a lot more to go to court with than very stale jokes.
Me? Well, last week we placed another order for more hp i4 blades as the competition still couldn't provide an option that could beat them. Sorry if that upsets those that rushed to the lifeboats simply because they thought an El Reg joke was a good basis for making purchasing decisions with.
Just curious - what is the workload for those blades?
I was poking around HP's site to see what the Itanium blades are up to, and it seems for any given physical size the HP Intel or AMD blades have more processing power and double the memory capacity.
1 x BL870c i4: max 24 Itanium Cores w/768GB in "two" full height blade slots (1 system)
1 x BL680c G7:max 40 Xeon cores w/2TB in "two" full height blades (1 system)
2 x BL685c G7: max 128(!) Opteron cores w//2TB in "two" full height blade slots (2 systems)
1 chassis:
Itanium: 96 cores @ 3TB
Intel: 160 cores @ 8TB
AMD: 512 cores @ 8TB
I'm sure that clock for clock Itanium is faster though it's got quite a ways to to make up for the massive numbers of cores on the x86 side. Nothing can be done about the relatively poor memory capacity of the Itanium blades though.
I was expecting a single width 4 socket Xeon blade but I guess HP doesn't make those at the moment.
I suppose if you needed an 8 socket blade there are no 8 socket x86 blades from HP anyway.. though most would be fine with a DL980 in that situation, when the costs are that big, I can't imagine a bit of extra space being a problem.
Of course x86 doesn't run HPUX..
"Just curious - what is the workload for those blades?...." Go ask the Fortune 1000, they just about all buy them, mainly for big database enviroments like billing systems, CRM, etc. If you made a mobile call in the last ten years then it's likely the charge for it was billed using an hp Itanium server.
"....I was poking around HP's site to see what the Itanium blades are up to...." I suggest you go read up on how hp can take the single BL860c i4 blade and group them in a modular manner to make larger blades - BL870c i4 is two, BL89c i4 is four. That means hp can make an eight-socket blade with octo-core Itaniums as one 64-core server image. They can aslo scale up with Superdome2 for even bigger implementations. And hp can do this with the top-end Itanium CPUs, unlike IBM which cannot put the most powerful Power CPUs in their blades, and can't modularise them in the same way hp can either, the best IBM can offer being thirty-two crippled Power cores in the 70x blades. And Sun/Oracle's attempts at blades are a subject for comedy, not IT.
".....I'm sure that clock for clock Itanium is faster though it's got quite a ways to to make up for the massive numbers of cores on the x86 side...." Go read up on the architecture. The Itanium has massively wider pipelines and many more registers, making it a much more flexible platform with more grunt for each cycle. It's a bit like comparing Sun Niagara with M-series, two different CPUs for different workloads.
"......Of course x86 doesn't run HPUX." Nor doe they run AIX or OS/400, and no-one wants to run Slowaris on them even though they can if they're feeling masochistic.
Back under your bridge, troll.
>>Back under your bridge, troll.
Matt, I read nothing into that post that came across as trolling, it looked like a genuine request for further information.
Calling him a troll in those circumstances doesn't strengthen the case you are trying to make.
Full disclosure: I have to suffer neither oracle nor hp anymore.
Fortune 1000 use bladed Itaniums for 'big database environments'? Not any I know. Most of them use servers such as p795, Oracle m-Series, or even, Integrity superdomes. They might use blades for testing environments maybe, but not really production. If you're a Fortune 1000, you can afford the proper stuff and get the benefits of consolidation, something completely lost with nPAR virtualisation.
Whether IBM can put their most powerful Power processor in a blade or not is completely irrelevant. What's the comparative performance of one of their blade Power processors and an Itanium processor? Also, maybe they have different marketing strategies. Maybe they could put their most powerful in there, but choose not to. It's a pretty small market segment that wants to be able to grow Itanium blades by linking them together into a single server. Once you get above a couple, most companies would simply go out and buy a bigger server and virtualise it. The only reason HP can do it, is because that's how they've chosen to build their Superdomes and therefore making it available in blades enclosures at the smaller level is easy. What's the throughput though?
Itanium and Xeon each have their own advantages. Itanium relies very heavily on being able to parallel workload to be able to use its architecture to the full. Not all workloads are like that. If Itanium was so good at heavy DB loads, why wouldn't Oracle simply get HP to make Exadata's for them using Itanium? Or, even make their own? One of the reasons, is that Itaniums cost a whole heap more than Xeon processors.........as in factors.
"Fortune 1000 use bladed Itaniums for 'big database environments'? Not any I know. Most of them use servers such as p795, Oracle m-Series, or even, Integrity superdomes."
Definitely. I don't understand why anyone ever used Unix blades for a large DB. It does away with one of the primary benefits of Unix, scale up with a huge single instance to dozens of sockets. Even if you don't need dozens of sockets, blade form factor is rarely the best solution. Unless your bottleneck is CPU (which it almost never is) and you only need a few sockets of that CPU, it doesn't make sense to consolidate so much CPU into such a small form factor with limited memory and I/O (which tend to be the bottlenecks). Blade systems actually use more power and cooling, floor space than large scale up systems, so it isn't about the environmentals. I am sure IBM thought that, with the I/O restraints of blades, people are never going to get enough load to the blade to fully utilize a 5.0 Ghz proc, so 3.8 Ghz will probably work just fine. HP's management loves blades primarily because they look cool in ads and give them an excuse to sell something new/more costly to rack users.
Also, the new BladeCenter, now called IBM Flex System, has fully clocked Power nodes including a native four way SMP.... not QPI linking of separate blades together. With Flex, you have a native four socket node with expanded I/O and memory, as compared to the previous generation, so high frequency procs probably makes more sense as you can get enough workload there to keep it busy.
I'm the IBM enthusiast... and even I know that teleco networks were always the domain of Sun Netra for the network nodes, probably moving away now... but there is still a ton of it out there. You cleverly mentioned "the billing", which has nothing to do with the complexity of teleco networks... it is a back office function. I know there are a few Superdomes out there, but they are going away. I happen to know, from personal connections, that one of the three major carriers in the US just replaced their Superdomes with Power 795s last month.
@AC.
The funny thing is, the court judgement doesn't actually help HP in the slightest. Whilst Oracle may be forced to support Itanium for longer, nobody wants to go with a company being forced to do something. Firstly, Oracle will do the minimum necessary. This means support for Itanium will come bottom of the pile. Secondly, Oracle will do everything it can to stop supporting Itanium or make it a non-starter for companies. Do you really want to be with a software vendor who doesn't want to support your hardware? The most costly and hardest bit to change is software, so get to another hardware supplier. At best, HP has bought a bit of time. If I was them, I'd be porting HP-UX to Xeon as fast as possible and start making Superdomes etc. out of Xeons. This would have all sorts of advantages to them, not least price. Unfortunately, after doing over their Tru64 and VMS customers in the past and the debacle over RISC and Alpha etc., customers aren't going to like another forced platform change again.
".....the court judgement doesn't actually help HP in the slightest....." he really funny thing is how desperate you and the other IBM trolls are to paint it that way. Gee, I wonder why - not!
"....Oracle will do the minimum necessary...." Actually, they will have to do what the contract says, i.e. provide versions of the Oracle software for Itanium in a timely manner. Any deviation from this means they get dragged back to court. This is what hp wanted, this is what us customers wanted, so hp and us customers won, and you IBM trolls just can't stomach that.
Enjoy!
"Actually, they will have to do what the contract says, i.e. provide versions of the Oracle software for Itanium in a timely manner. Any deviation from this means they get dragged back to court. This is what hp wanted, this is what us customers wanted, so hp and us customers won, and you IBM trolls just can't stomach that."
Yes, but if you bothered reading and UNDERSTANDING my point, there is no requirement for the Itanium to have anything beyond the features it has today and nothing to stop them leaving the Itanium version standing. I've just shown in another reply that Exadata has unique features which means they can quite happily have different versions with different feature sets. So, all they need to do, is create a new version (i.e. increment the number), make a couple of changes (not the same as every other version) and carry on support. Fine. But, very soon, the version of Oracle will be a long way behind everyone elses. At that point, the quality of the hardware becomes irrelevant.
Would you really force someone to maintain your car under pain of a court settlement and then wonder why they do the bare minimum rather than a quality job?
".....Re: Ill-educated Mike pm
".....if you bothered reading and UNDERSTANDING my point.....". Point? You don't have a point, all you have is an irrational hatred of Itanium, probably based on fear because it is a UNIX platform and therefore one of the ones gutting yout precious mainframe bizz!
"..... there is no requirement for the Itanium to have anything beyond the features it has today.....". And you didn't bother to actually go read any of the background on the Oracle-hp agreement, even when I pointed out in a previous post that there is already an existing agreement, re-inforced by the "business as usual" contract, that ties Oracle to producing new features and software versions of their products now on Itanium at the same time as they do for other OS such as Slowaris. In essence, the only way Oracle can stop adding features to those software lines is to stop making those products lines - such as Oracle DB, RAC, Application Server - for ALL other OS, including mainframe (for which they already do not make many of the software they sell on Itanium) and AIX. Larry would have to commit business suicide spite hp, otherwise hp will simply drag him back into court and extract more cash every time he dicks around. You really need to try and get yourself out of the mainframe bubble so you can actually see what is happening in the real world, maybe a Zimmerframe would help?
"......Would you really force someone to maintain your car under pain of a court settlement and then wonder why they do the bare minimum rather than a quality job?" Which is a stupid statement as hp have a contract which ties Oracle to producing versions and support to a set level. IBM have nothing of the kind. Really, READ THE NEWS!
"Point? You don't have a point, all you have is an irrational hatred of Itanium, probably based on fear because it is a UNIX platform and therefore one of the ones gutting yout precious mainframe bizz!"
Mmmm. Not following the thread again. If you read backwards, you'll find that I said I USED TO WORK on mainframe, but I now WORK ON UNIX systems. So, far from being fearful, I'd probably like it to do better!! I have some residual responsibility for mainframe, but the vast majority of my work is on Unix. So, try reading backwards and following the thread rather than making your own entries up.
"And you didn't bother to actually go read any of the background on the Oracle-hp agreement, even when I pointed out in a previous post that there is already an existing agreement, re-inforced by the "business as usual" contract, that ties Oracle to producing new features and software versions of their products now on Itanium at the same time as they do for other OS such as Slowaris. In essence, the only way Oracle can stop adding features to those software lines is to stop making those products lines - such as Oracle DB, RAC, Application Server - for ALL other OS, including mainframe (for which they already do not make many of the software they sell on Itanium) and AIX. Larry would have to commit business suicide spite hp, otherwise hp will simply drag him back into court and extract more cash every time he dicks around. You really need to try and get yourself out of the mainframe bubble so you can actually see what is happening in the real world, maybe a Zimmerframe would help?"
Afraid you're reading the agreements etc. wrong. Business as usual doesn't mean anything of this sort and doesn't tie Oracle to any such thing. If it did, how come the two Exadata only features (which were both available before the HP/Oracle spat) aren't available on the Itanium version or indeed, any other version? Because your interpretation above is wholesale wrong. BAU has never and will never mean being able to insist on all features in all versions. If you really want to prove your point, perhaps you could point me at an article that tells me when Hybrid Columnar Compression and Smart Flash Cache are going to be made available on any platform apart from Exadata. You MAY possibly find it for Sparc, but certainly nothing else. And that's only because an Exadata equivalent is/has (not sure of the exact date) being launched that uses Sparc for the database nodes at least. Storage nodes stay as x86 I believe.
I do hope you're not trying to persuade your management you haven't made an epic fail as you'll be looking for a new job soon..........that's if you currently have one!!
".... I said I USED TO WORK on mainframe, but I now WORK ON UNIX systems...." Sorry, but I'm just having a very hard time believing someone that knows so little outside the mainframe bubble - well, going by your posts, NOTHING outside the mainframe bubble - is employed at all! I assume the company decided it was cheaper to keep you on until retirement rather than make you redundant when they replaced your mainframe with something relevant.
".....Business as usual doesn't mean anything of this sort and doesn't tie Oracle to any such thing....." <Sigh>. I suppose the news probably didn't filter through to the alternate reality of mainframers, but hp and Oracle signed a co-development agreement years ago before the Sun purchase. If you go to the Oracle site at Reading in the UK you can see the ETC building which hp funded and equipped with hp systems for hp and Oracle to use for joint POCs, etc. I know because I was one of the customers that made use of the facility. Prior to that, Oracle did all their development on Sun Slowaris, meaning that new versions of products like Oracle DB were released first on Slowaris, then developed for hp-ux. The co-development agreement meant Oracle was tied to releasing versions of products on Slowaris and hp-ux at the same time. The "business as usual" contract ties Oracle to the same concurrent release schedules for all products currently available on hp-ux for Itanium. Oh, and IBM had no such agreement with Oracle, another advantage of hp-ux.
".....the two Exadata only features ....." Exadata is an APPLIANCE, not a version of Oracle DB! FFS, go get a dictionary and LEARN something.
"Sorry, but I'm just having a very hard time believing someone that knows so little outside the mainframe bubble - well, going by your posts, NOTHING outside the mainframe bubble - is employed at all! I assume the company decided it was cheaper to keep you on until retirement rather than make you redundant when they replaced your mainframe with something relevant."
Abuse as usual for no answer. I was actually doing mainframe work, when a company recruited me for Unix work, which at the time, I didn't know a lot about. Fortunately, they realise that some people learn very quickly and I've now been doing it for them for some time.
"I suppose the news probably didn't filter through to the alternate reality of mainframers, but hp and Oracle signed a co-development agreement years ago before the Sun purchase. If you go to the Oracle site at Reading in the UK you can see the ETC building which hp funded and equipped with hp systems for hp and Oracle to use for joint POCs, etc. I know because I was one of the customers that made use of the facility. Prior to that, Oracle did all their development on Sun Slowaris, meaning that new versions of products like Oracle DB were released first on Slowaris, then developed for hp-ux. The co-development agreement meant Oracle was tied to releasing versions of products on Slowaris and hp-ux at the same time. The "business as usual" contract ties Oracle to the same concurrent release schedules for all products currently available on hp-ux for Itanium. Oh, and IBM had no such agreement with Oracle, another advantage of hp-ux."
For many years, the primary development platform for Oracle has been Oracle Linux, not Solaris. As you're so knowledgeable, could you tell me when that agreement runs to? Think you might find that agreements tend to have an end date and aren't valid after that.
"Abuse as usual for no answer...." Plenty of answers, it just seems I need to try using very short and simple words for you to understand. It's gettign hard as I don't think The Reg will allow me to attach crayon diagrams, which seems about your intellectual level.
"....which at the time, I didn't know a lot about...." Going by your posts, not much seems to have changed there. It also explains why you know SFA about the Wintel/Lintel environment.
".....For many years, the primary development platform for Oracle has been Oracle Linux...." On hp Proliants, actually. Oracle try to hide the fact they didn't even move their labs over to Oracle x64 servers until people started realising how much hp kit they were still running. But please do try denying they still develop for Slowaris, it would be very funny to see what frothing claptrap you can come up with on that one!
"....Think you might find that agreements tend to have an end date...." In another amazing bit of Oracle contract ineptness (it was agreed in the days when Oracle viewed themselves as staying as just software) there was no end date to the co-operative agreement regarding version parity. But you didn't even know about that agreement, so I'm not surprised you didn't know anything about the terms. Fail!
> Itanium: 96 cores
errr nice little piece of arithmetic there.
You take 2 8core CPUs per blade and you can fit 8 of these blades into an enclosure.
so, 2*8*8 == 96. :-) cool I never knew that.
Of course with the BL8[679]0 boxes you can decide on whether your 8 blades should be 8 separate systems, or perhaps 4 slightly bigger one, if the that suits you needs or then again perhaps you'd like something a little bigger and so have it configured as 2 64way boxes. Then again, you can also have any combination, just by changing the handles on the front. They don't have that functionality on the ProLiants... yet.
You're right about the RAM figures. The ProLiants Gen8 boxes can house more RAM.
BTW, a DL980 doesn't quite make the OLTP figures for a circa 2007 SuperDome (not talking about an SD2).
> Of course x86 doesn't run HPUX..
Now me, I love Linux, but there are some things that HP-UX is just really good at.
Really getting hit hard this year it seems :-(
Kinda feel some sympathy towards em, but at the same time I have no sympathy for those that use the IA-64 architecture, it was just a 'phase' architectures were going through...
Now, Get with the times! (No not the news corp!)
With a better architecture, database,middleware and apps! *facepalm*
If Oracle has any interest in acquiring HP, this would be an opportune time. I don't think that they have any interest in HP, but, if they are considering it anyway, they might be pushed over the edge if it looks like they are going to have to cut HP a check for a billion or so. They could acquire them, spin off PCs and print... possibly some other stuff, and cut a check to themselves. Highly unlikely, but stranger things.
First, I have to say I was with HP until 1997, when PA RISC was still going very well. These machines were undestructible by accident. You had to flood them with water or run over the box with a tank. They did simply not crash like the Win 3.x crap we also had at HP then. I got them crashing only by means of "ping of death" or "crashme". In normal operation the OS never died. You got "bus error" from faulty software and that was it. HP CC was excellent it optomizing scientific code. No need to hand-optimize with -O2. PA RISC hardware was generally leading ALL processors and the only serious competitor was Alpha of DEC. x86 was nowhere near PA RISC performance.
I don't know much internals after 1997, I only heard Itanium was consuming massive amounts of power and had a consequential cooling problem. Which true HP engineers would have found a solution for, no doubt. Water coooling, if required.
But the VLIW (which Itanium is the latest incarnation of) concept is very interesting and promising, especially when Loop Unrolling is possible. Itanium did have very impressive floating point performance in the past and I do also think that the strengths of VLIW could be exposed if the programmer is aware of them.
http://en.wikipedia.org/wiki/Very_long_instruction_word
"The VLIW architecture is growing in popularity, particularly in the embedded market"
So when commenters with their shallow hate come out and try to smear HP for "Itanic", I do think this is a very low point for IT people. Or maybe the shitlobbers are just professional propaganda whores ?
Very much like Boeing with the 787, HP tried to do something very innovative in several ways. First, technology innovation with VLIW. Secondly "outsourcing" was the big rage in the 90s and senior HP management thought it would be a good idea to hand over CPU design expertise to Intel and to hand over compiler development to some indian outfit. Operating systems and databases could be bought from MS and Oracle. HP leadership thought they could cut expensive American engineers out of the equation and still make massive revenue and profits, just as they did with PA RISC and HP9000 and HP3000.
In hindsight, we all know that it was wrong for HP to treat their crown jewels (HP-UX, MPE, PA RISC, Allbase/SQL) and their creators with that amount of disresepect. But then it was 100% OK to dream the pinko-liberal new-age dream of "let's rake in tons of cash by using third-world labour". After all, "programmers are all the same, why should I use the expensive, experienced and slightly disobedient white guy if I can get a brown guy at 1/10th of the cost ?"
Still, I do think it is irrational and nasty to badmouth innovative technology like VLIW. SPARC is not bringing anything better on the table and if HP had retained those expensive, experienced Americans and their own design and manufacturing capability, they would either have a competitive PA RISC or VLIW machine by now. They were leaders in the 90s with two teams (PA RISC and Alpha). Intel was far, far from their performance levels.
Today, Itanium is, very much like the IBM S/360/390 architecture, at the heart of hundreds of billions of dollars of business transactions every year. These systems simply work, instead of crapping out for random, stupid reasons such as "too cheap capacitors". They are business workhorses and they often run the Oracle database management system. HP was in fact a major part of the Oracle salesforce, as HP was falling over each other in selling/implementing "hot", external stuff such as Windows, Oracle and SAP.
It is just a tiny bit of justice that Oracle has been ordered to continue to support the Itanium architecture as Oracle would not be where they are without the help of thousands of HP engineers and salesmen.
"the pinko-liberal new-age dream"
Forgive me if my understanding of US politics is not up there, but I thought the whole outsourcing idea was very much a Republican/right-wing system to profit the few at the top (i.e. shareholders / CEO), rather than the workers? (usually the "pinko" term is applied to what would be middle/left in the UK politics).
As for VLWI based machines, I used some TI DSP based on that some years ago and found that the compiler technology was piss-poor in delivering anything like the promised performance. It needed tedious hand-optimisation of C code and/or the "sorry, life if too short" resort of assembler (after learning the architecture inside-out) to get there.
In my humble opinion, dropping the Alpha was possible HP's worst move of all if they had any intention of being in the non-commodity hardware business.
"In my humble opinion, dropping the Alpha was possible HP's worst move of all if they had any intention of being in the non-commodity hardware business."
Pretty much anyone who had any experience of Alpha would agree. I programmed Alpha and had a few as home machines running 64 bit Linux at a time when the FP performance for a 21264 box was 8x that of the fastest Intel processor and for integer work, having 32 full 64 bit registers, register renaming, instruction reordering and four independent integer pipelines with two floating point all lead to a very very fast CPU which you could write assembly for almost like a high level language. For years various wags had been claiming Intel would roll into the 64 bit market and take over with Merced. Well, that didn't happen and while HP had been busy working with Intel to develop it and kill their own PA RISC, Compaq bought DEC and got the Alpha business and continued to develop it. When HP bought Compaq, they killed Alpha because it competed with their own plans. That was a mistake. They took the best performing, most scalable CPU architecture on the market and killed it when it was the speed king and they replaced it with an untried architecture which relied on very smart compilers to get any performance (anyone else remember the dreadful i860 from Intel?) and of course that didn't work. Many Alpha engineers hit the market, quite a few got sucked into AMD and the Opteron was the result and it has taken years for Intel to get properly competitive with those and they could only do it because of their strength of market position and deep pockets.
Itanium is the CPU equivalent of Windows 8. It deserved to fail. If only HP could suck it up and relaunch Alpha but it is likely much too late so we get the modern day x86_64 architecture which, while compatible with ia32, is horrible to code for (limited registers, bolt on vector operations that you have to jump through hoops to use and so on) but at least they're cheap.....
This post has been deleted by its author
This post has been deleted by its author
"Forgive me if my understanding of US politics is not up there, but I thought the whole outsourcing idea was very much a Republican/right-wing system to profit the few at the top (i.e. shareholders / CEO), rather than the workers?"
He is right... the "globalization" or "free trade" initiatives, offshoring being an adjunct of free trade agreements, were nearly all Democratic goals. The Democrats are basically left wing on social issues and right wing on economic issues, not as right wing as the Republicans in most cases but still right of center from a European perspective. Clinton did NAFTA, WTO, FTAA, etc. The groundwork was almost entirely set by Clinton.
But fortunately still quite an interesting discussion - reminiscent of the Unix wars of old.
My view on VLIW is coloured by having done development on an Apollo DN10000 - performance was utterly dependent on whether one' s coding style suited the compiler, and compiler bugs abounded. Not to mention horrendous compile times... The HP 9000/730 that replaced it was like a breath of fresh air, but we did miss the Apollo Domain OS.
IMO the Itanium approach was definitely wrong, an Intel complification of something that started as a simple elegant and compiler-friendly architecture. What HP *should* have done is to port HP-UX to the Alpha architecture... although I never got to play with one, we used to follow computational benchmarks and Alpha was always ahead of the pack. If only they had paid Intel to fab the PA-RISC processors and thereby gained all the Intel process advantages without the architectural dog's breakfast that is a hallmark of Intel's efforts in this field.
BTW across a variety of PA-RISC machines used in those years, I remember only one instance of an OS crash, which has as its root cause dodgy firmware on an LAN card (VG100, remember that?)
In terms of performance, I don't think there was much to chose between PA-Risc and Alpha. Back when PA was HP's main processor, HP and DEC used to trade the SPEC Mark crown between them, based on who's chip had been updated most recently. There was rarely much between them.
Alpha was effectively killed off by Compaq, as part of settlement of a legal fight between DEC and Intel, the design team of Alpha was sold... and they didn't like it. So most of the team left.
Ownership of the specification passed to HP with the merger between HP & Compaq, but the real prize was the team of engineers and they'd gone already.
Itanium started life as PA3 and I first heard of it about a month after the development project for 64bit PA (PA2) was kicked off.
Sadly the first version of Itanium, Merced, was about 3 years late in coming to the market. The chances are that if it had come out when it was planned, it would have taken over as planned and we wouldn't be having this discussion. Intel wouldn't have spent the $Billions on rescuing the brain dead x86 architecture and some of that money would have been spent on make the compilers of Itanium work better.
In terms of Fabbing PA processors. HP stopped fabbing them themselves donkey's years ago. The first PA processors (840/930) were made out of discreet TTL chips. After that HP used various fabs, PA8000 and PA8200 were fabbed at Intel if I remember rightly. PA8500 was fabbed at IBM.
"It is just a tiny bit of justice that Oracle has been ordered to continue to support the Itanium architecture as Oracle would not be where they are without the help of thousands of HP engineers and salesmen."
It is a bit of a stretch to say Oracle would not be where they are without HP. HP's preference for Oracle was not done out of charity for Oracle. When they had any say over the apps running on their hardware, which was rarely... as with any server provider, they went with Oracle DB because the obvious alternative, especially after Sybase fell off, was IBM DB2 for enterprise level DB. HP didn't want IBM in their accounts, so they went with Oracle for their own benefit, not Oracle's benefit. When someone wanted MS SQL or DB2 or SAP instead of Oracle ERP, HP doesn't say "no, we are special partners with Oracle" or even "you should really take a look at Oracle instead" they were and are happy to sell Oracle's competitors' software.
HP-UX was a relatively late comer to Oracle. DEC and IBM had been working with Oracle for a decade before they created any Unix port. If any other IT company was essential to Oracle's success, it was IBM with early access to System R and giving them the idea for RDBMS in the first place. If a second partner was critical, it was probably DEC with PDP-11 and VAX as the first Oracle platform. If a third partner was critical to Oracle, it was Sun in the .com boom years as the go to DB for web apps. For most of Oracle's rise, Sun SPARC - Solaris was their go to hardware partner.
Why didn't HP push Allbase/SQL ? At least in theory it could have been where DB/2 is now. It could actually be the basis of an HP software business, very much like DB/2 is a cornerstone of the IBM software activities.
And please, don't tell me Oracle and/or DB/2 were superior then. Maybe on the glossy marketing sheets, but in reality the internals of Oracle and DB/2 look very nasty. The Oracle 8 listener could be crashed using dangerous hacker tools like "telnet ora-server 1542;some random typing on the keyboard". DB/2 is a nightmare if you expect to "run it out of the box without an IBM customer engineer holding your hands".
The problem of HP always was that they took a cynical attitude towards their own products and technologies. HP management never really fought for their products like Gates, Ballmer and Ellison did. Instead they were anxious to "sell leading competitor products with our hardware".
Finally they convinced themselves that they didn't need any of their own CPUs or operating systems any more. Can you say "emperor's new clothes" ?
But it was a great time for HP in the 80s and 90s.
".....The problem of HP always was that they took a cynical attitude towards their own products and technologies....." Well, yes and no. As I understand it, having spoken to peeps at both Oracle, hp and Sun (yes, even them!), hp decided to look at what the target market were actually DOING and then go from there. IBM did not. What hp saw was that Sun was king and mainly because it was a prime Oracle partner - just about every customer they talked to wanted Oracle DB rather than any other database, but were happy to talk about a different platform. IBM simply plugged along the lines of "buy a mainframe - no, then buy a P-series with AIX and DB2". What hp did was look at Allbase, realised that pushing that against the Sun-Oracle partnership was going to be just as silly as IBM's insistance on DB2, so instead decided to beat Sun at their own game. Whilst IBM was vainly trying to replace Oracle, hp set about replacing SPARC and Slowaris underneath the Oracle DB. Part of that process was cuddling up to Oracle, which was pretty amazing considering they were also cuddling up to Microsoft, and later Red Hat, SAP, and just about anyone else that mattered. Even IBM Software - more licences sold on hp servers than IBM ones for years due to the Proliant dominance of x64.
Thanks for the info about fabbing... still trying to square that with my recolections of the time, when HP's progress seemed to slow down after the 9000/735 - their processors seemed to be getting a lot hotter for comparatively small performance bumps.
In response to the previous post, the i860 did come to mind as the poster child for Intel's knack of producing architectural stinkers.
Great system, but the last in the line. The 735/125 packed a PA7150 CPU.
This was followed by the J class workstation and K class servers. The J should have given you quite a jump in real world performance. These boxes packed the PA7200. Same clock speed, very limited super scalar capability, really neat fully associative assist cache to avoid mapping contention in the direct mapped L1 cache.
Big jump in performance after that with the PA8000, HP's first 64bit processor, performance was down by about 20% on its capability due to the limited availability of cache RAM chips able to keep up with the CPU.
PA8200 came next, at 240MHz it matched the Alpha 21164 @ 666Mhz, but the PA processors had had out of order and speculative execution since the PA8000, which the Alpha didn't get it till the 21264. Both the 21264 and PA8XXX chips had extensive 4 issue super scalar functionality.
The PA8200 was the last of the PA-Risc chips to make use of external cache. The clock speeds had become too high to be able to continue with that strategy, as PA8000 had shown. So the PA8500 switched to on chip cache, faster but smaller, they could only do 1.5MB of cache (that's L1 cache, 1MB data, 512K inst).
Minor changes in the PA8600, before the PA8700 came in taking the caches and TLB up by 50%.
Sorry can't remember the changes for the PA8700+.
PA8800 and PA8900 are the run out dual core chips, these wimp out and implement L2 cache (32MB and 64MB).
The i860 doesn't really have any place in Itanium story, at least not that I know of. The i960 does, it was the cancellation of the i960 CPU much beloved of HP LaserJet's that drove HP and Intel into bed together on the CPU market place.
When Compaq sold all the Alpha IP to Intel in 2001 a bunch of the engineers left and joined AMD producing the Hammer. This killed the Itanium workstation business over night,
Lots of blame to go around here; just don't include the kinds of problems HP has had keeping up with Intel XEON and IBM Power CPUs to this issue.
Basically I see Oracle's strategy (i.e., Larry and Mark) as creating a new "Oracle Platform Short List" for customers. The old one was Intel (IBM, HP, Dell and others on Windows, Linux), IBM Power (AIX, Linux for Power), and HP Itanium (HP-UX, Linux for I64). Oracle was able to buy Sun Microsystems at a steep discount because Sun didn't have anything to offset revenue being bled off by Intel x64 Windows and Linux boxen, unlike their IBM and HP competition. But once Sun became part of Oracle, cutting off HP-UX and Itanium from the Oracle base would leave space for Solaris on SPARC or XEON x64 as the other choice on the new "Oracle Platform Short List" . Pretty simple really - anti competitive behavior to be sure, but all's fair in love and hardware, eh? I can therefore see the judge's point - no reason to not offer Oracle on HP-UX. The issue is who pays for it?
Remember - HP-UX requires specially optimized code for Itanium; code meant for the VLIW "EPIC" instruction set at the core of Itanium. So Oracle has a point as well - they can use the same code base (with tweaks) for Windows Server, AIX, Linux, and Solaris; but to maintain support for HP-UX, they need to keep maintain a separate set of source code for Itanium to perform properly. If they do this the right way, HP-UX stays competitive with AIX and Windows/Linux; but if Oracle doesn't do the work to get the most out of its Itanium software variants, then performance will suffer versus HP's competition, and even against HP's own Windows and Linux XEON x64 systems. If I am Oracle, why should I have to incur additional cost for one specific HW platform ( and one that I compete with)?
I think the answer then is straightforward. Oracle has to pay HP for damages done, but then it should also tell HP that the latter henceforth has to bear the differential cost of "optimizing" Oracle's DBMS code and other software for Itanium- based servers, because of their _special_ nature. HP would then face a choice between taking the money from the lawsuit and seeing HP-UX sales suffer further due to poor Oracle performance for Itanium systems, or making nice with Larry & Mark. In other words, I believe the operative phrase here is "Pyrrhic Victory".
YAAC
AC: I don't know whether you've ever coded, but I don't think Oracle have ever done any specific optimizations for Itanium. That is what the compilers do. Sure if you want to hand code in assembly then an Itanium is a very different beast to code for. But for high level languages, nope, you don't need to do anything special. What Itanium does offer are the possibility to analyze exactly where in your code you're suffering from cache misses, TLB faults ... etc. but you'd be suffering just as badly on any other platforms, they just can always tell you why.
What sorts of things do you feel you have to code differently just because its a VLIW system?
I notice you haven't taken up my bet yet on whether mainframe or Itanium will die out first................
Are you going to provide the links for Exadata only features appearing on any other version of Oracle DB for another platform......................
I'm sure the silence will be deafening, as you'll loose on both!!
I've chosen Paris because she's a clueless, brainless person who can't survive without people around her to help. Reminds me of someone..........
"I notice you haven't taken up my bet yet on whether mainframe or Itanium will die out first....." There's two reasons for that - firstly, I simply worry that you're so old you will join the other dinosaurs in extinction before the bet is decided; and secondly, I try to avoid the mentally-impaired and deluded.
".... Are you going to provide the links for Exadata only features appearing on any other version of Oracle DB for another platform....." I see you're still missing out on the small fact Exadata is an APPLIANCE and not just a version of the Oracle DB. Well, at least your monumental level of fail is consistently perfect.
".......I'm sure the silence will be deafening.....". Already pointed out, Exadata is an appliance, you just need to go find an adult to explain that word to you.
"......as you'll loose on both!!...." See, this is your problem - well, one of your problems - you want to declare a victory before the event. Do you have time travel? No, so you are just urinating into the breeze and claiming that being wet makes you the winner, whilst everyone else just wishes you'd go somewhere else. Fail!
".....I've chosen Paris because she's a clueless, brainless person who can't survive without people around her to help....." Strangely fitting given the vacuous nature of your posts and their complete divorce from the realities of modern-day IT. St Paris, Patron Saint of Mainframers - the best suggestion you've made in all your posts! Unfortunately for you, going on the evidence presented by your dribbling a I'd have to say that it is a massive insult to Paris to associate her with someone as deluded and clueless as your own words prove.
"There's two reasons for that - firstly, I simply worry that you're so old you will join the other dinosaurs in extinction before the bet is decided; and secondly, I try to avoid the mentally-impaired and deluded."
I take it that's a no then. Obviously, you don't have any faith in your own comments!!
P.S.
Exadata runs Oracle DB in just same way Integrity does. Calling it an appliance (of as I said earlier, 'Engineered system', which is what Oracle actually call it) is simply a marketing stunt.
"....Obviously, you don't have any faith in your own comments!!...." As pointed out before, the really worrying thing is yo actually believe the male bovine manure you spout.
"....Exadata runs Oracle DB in just same way Integrity does...." <Sigh> Just exposing your lack of knowledge there again. Exadata is a carefully tuned stack of servers, storage and Oracle software. You could build a similar stack on Itanium using Oracle DB and hp Integrity but there are some software bits, mostly management, that are unique to the Exadata product. However, you can build a superior system if you use hp's Matrix product, especially as that allows you to build a solution that matches the individual company's requirement, around mixed environments of Linux, Windows and hp-ux, and using non-Oracle software and applications if required, and can be scaled up and down, compared to Oracle's one-size-fits-all and Oracle-software/hardware-only approach. I'd try and explain Matrix to you but I fear you are really just too obtuse to get it, and there are people at hp actually getting paid to sell it, so I suggest contacting them and seeing if they have the patience to deal with someone as "special" as you.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
