With the quad-core "Tukwila" Itanium processors coming out sometime around the middle of this year, the Itanium Solutions Alliance, the consortium of hardware and software vendors that peddle Itanium-based products, has dusted off the Itanium drum and begun banging on it. If there is one thing that you can probably say about …
Itanium = HP
When (not if) Itanium dies because Intel HQ realise they get better return on investment by investing in x86, HP could port their enterprise OSes (HP/UX, NonStop, VMS) to the current incarnation of AMD64, which is a beast already well known to them. The OSes have all been ported before, the technical issues are (subject to the knowledge involved not having been lost) known, all that's needed is the commercial will.
What kind of idiot would buy an already-obsolescent Itanium box to run SLES11 on, when SLES and a whole load more runs just as happily on real "industry standard servers", and with a rather broader choice of Linux/x86 apps to go with it?
Or maybe, when Itanium goes EOL, HP give up on "enterprise" altogether and survive on a combination of printing, Pavilion, and Proliant (and maybe Procurve).
"Embedded x64 chips"? Er, I don't think so. Can anyone explain? Afaict, depending on sector, Power and ARM have that market sewn up between them for the foreseeable future.
Nobody can stop the X86 train :-)
not even Intel ....
sparc vs itanium
Get Real -- If nobody bought any sparc chips after this year Itanium would still be outnumbered for years to come.
"If Itanium wants to outlive Power, what Intel and HP really need to pray for is that the x64 chip takes away all those game console deals from Big Blue. They need embedded x64 chips, like the Atom, to take off as an alternative to embedded PowerPC chips, too. IBM better stop and tie its sneakers if it wants to outrun that grizzly."
Huh? Game consoles do not hit me as an that important to an architecture that is commonly used in the automotive industry for example. That Grizzly is a bit pear-shaped and the probability that it will reach Itanium before Power even feels the body heat is about 1. We need some numbers here though.
Itanium = Dead
Whether SPARC is dead or not does not matter... Oracle will milk SPARC for another 5 to 10 years as the current customer base of SPARC is 10 times that of Itanium and well worth the effort for Oracle... Itanium on the other hand makes no money, relatively speaking, for Intel. Intel want's Itanium dead and the only thing that keeps it going is HP's bribes to keep it afloat. Intel's future is X64 and as a result, X64 is HP's future.
Ready, set, go MB!
Itanium Also = Windows and SQL Server
If you want to run a very large SQL Server-based database then Itanium is really worth considering. The only alternatives are units like IBM pSeries and similar. You can run Windows Enterprise Itanium Edition and SQL Server Enterprise Itanium Edition. The overall cost is much less than that of a Unix and Oracle configuration. The cost of SQL Server software is a fraction of the cost of Oracle for a similar configuration - up to 95% cheaper.
Admittedly the only real Itanium-baser server vendor is HP with the Integrity and Superdome series. But banish your prejudices when considering large scale database solutions for high-volume transaction processing.
Note that there are architectural limitations in Windows Server Enterprise Edition where a maximum of 32 processors/64 cores can be supported.
"If Itanium wants to outlive Power, what Intel and HP really need to pray for is that the x64 chip takes away all those game console deals from Big Blue."
*cough* You mean the Larrabee chip
Embedded x64 chips
refers to Intel Atom. Some, but not all, of the models implement the x64 instruction set. ARM, Power and MIPS have better performance per watt and are well entrenched; so, Intel is facing a significant uphill battle. Intel Atom is more likely to (continue to) find use in netbooks, nettops, and thin clients than in "embedded" applications.
For *very* large databases Nonstop on Itanium is worth considering. Nonstop supports running a single Nonstop SQL database over 4000 cpus (8000 cores) with each cpu having 48GB non shared memory (Nonstop is a parallel shared nothing architecture). Not that anyone would ever need to go that big, but running several hundred cpus is realistic for some applications. For Microsoft SQL Server I can't see much benefit in going the Itanium route over x86.
I think HP will change course
Let's face it. The Itanic is sinking. HP would be insane not to port HP-UX to x64. The Nehalem is a killer architecture and Tukwila is yesterday's news long overdue. The future of SPARC is largely up to Fujitsu. As I see things though once the eight core Nehalem EX comes out a 64 socket server running that chip will probably blow even a Cray off the planet. And if IBM are smart they will be the ones to sell it. They don't even need to port AIX to it. They already have Linux and WebSphere. IBM can still keep their increasingly POWER based i and z series markets, but we all know the killer app of today is the front end web not the DBMS, which IBM also has covered. So they can let Microsoft deal with the OS. But so does HP with their alliance with Oracle. So it really depends on who beats the other to the punch. In my view HP's at least publically stated unwillingness to port HP-UX to Nehalem Xeon servers is a mistake. Of course HP has often concealed business strategies so they could spring a new version of HP-UX or even Linux on Nehalem at the last minute just to upset the market. A company that produces a pink digital camera branded by Gwen Stefani is capable of anything.
Intel Atom = Windows engine?
"Intel Atom is more likely to (continue to) find use in netbooks, nettops, and thin clients than in "embedded" applications."
Maybe in a few years I'll be able to buy something that's as practical as my Psions or my HP Jornada 720 were a few years ago. But it seems unlikely while it's got Intel Inside (tm).
In the meantime, MS and their lapdogs will use their usual, er, strongarm tactics with their Most Valued Partner system builders, by using discounts to ensure every volume product ships with Windows not A.N.Other OS, even if Windows isn't what the end user wants?
Yeah, well, after what Bill did to Bob over Shark you'd accuse him of strongarming too:
Meanwhile, Itanium today is in the same downward spiral as Palmer said Alpha was in back then - low sales volume = infeasibly high cost per chip just to recoup hw+sw development costs, never mind manufacturing cost plus profit. CSI (now Quickpath) was going to fix the hardware part of that cost by permitting compatible sockets for Xeon and Itanium, except it turns out that just like Blue Labour cabinet ministers, Intel never really said what we thought we heard.
Proof if you need it
Cost of second hand Dell 3250 dual Itanium server on ebay = £50.
Rackmount rails for it = £95.
Yeah, it says something when the rails are worth more than the server.
It's dead, Jim.
PS: Anyone want a Dell 3250....!
/Paris, because she's off the rails too
Re: I think HP will change course
"They don't even need to port AIX to it. They already have Linux and WebSphere."
I agree with most of what you say, but this line... Linux chokes at 16 cores now and it will take at least another 5 years to reach 64 cores. It will take another 10 years to work out all the little issues that go along with parallel computing. It IS rocket science.
Please don't bring up clusters... That only shows your ignorance. We're talking single images here. You can't run everything in a cluster. As it turns out, very little in fact. I know, I know, web servers are very well adapted to large clusters of Linux boxes, but again, that's not what I'm talking about!!!
Intel to blame
Intel have really screwed HP by slipping Tukwilla yet again. Right now the only Nehalem systems you can buy are limited to 2 sockets. If Tukwila had shipped in Feb of this year (and lets not forget even that date is over a year later than the original road map) then HP would have been able to pitch Itanium to any enterprise that was in the market for 4-core chips with 4+ sockets, without any competition from Intel's x86_64 line. That window of opportunity is shrinking rapidly and HP must be gnashing their teeth at the 100's of millions of lost revenue as a consequence.
2010 is going to be the year that really cripples Itanium. As you said, the 8-core Nehalem on 4+ socket systems is going to eat it alive. And don't expect that to be the end of the story, as 2010 is when AMD is due to ship its Istanbul based 12-core Magny-Cours processor, quickly followed with a 16-core version in 2011. So Intel will be forced to up the Nehalem ante very quickly to stay competitive. There just isn't enough of a market with Itanium for Intel to justify the investment required for Itanium to keep up with this. Especially as the bulk of Itanium sales in HP seems to be coming mostly from replacing legacy PA-RISC, not from new business. So even that growth has a limited shelf life.
It's a shame: Itanium is a really capable processor. But I believe the server market today is way too cut throat and moves too fast for it to survive more than a few years. If HP-UX were a compelling differentiator then that could make a difference. But Redhat Linux has matured quickly and is more than a match for that now too. I fully expect Redhat on Xeon 7500's in 2010 to give HP-UX on Itanium a very bloody nose. Quite how HP are going to respond to this threat to their Itanium business I have no idea.
If itanium fails
It will be much harder for HP enable their customers easily to migrate their code from Itanium to x64 as Itanium runs as big endian at least on one of HP's platforms.
I'd be surprised if Itanium outlives SPARC.
The real battle ...
... is between x86 and ARM. ARM is moving up from primarily embedded use to use in low-end servers, netbooks and will eventually move into high-end servers and desktop computers. At the same time, Intel is pushing x86 into hand-held devices and embedded systems.
While the market for desktop PCs and laptops is strongly tied to x86+Windows (with Apple being the main exception), neither embedded systems nor servers are so strongly tied to specific processors, so these markets are more open, and processors must compete on real values rather than by being the only one to run a certain software platform. So far, Intel (with x86, not Itanium) seems best in the high-end server market and ARM sits quite strongly in the embedded market, but if either one can make a processor that has the right combination of low price, low power use and high processing power to make it fit the other market, things can change.
Personally, I think ARM has more chance of getting into mid to high-end servers than x86 has of getting into embedded systems, but only time will tell for sure.
The rest of the processors: PPC, Sparc, MIPS, Itanium, etc. don't stand a chance of gaining a majority in either market -- they may retain sufficient market shares to survive and possibly even grow a bit, but they will never dominate the way ARM and x86 do.
If another processor architecture is to gain a significant share, it will have to be something new that is designed for massive parallelism, and it will run entirely new software: new operating systems, new programming languages, new applications. It will take at least a decade for this to happen, but I can't see us being stuck with inherently sequential software forever when the hardware is getting more and more parallel.
Re: I think HP will change course
"Linux chokes at 16 cores now and it will take at least another 5 years to reach 64 cores"
Dear Anonymous idiot. What a load of FUDy old rot. Get with the times. Linux scales happily way beyond that point. Don't believe me, let me point you (with an ironic twist) to the Itanium Solutions Alliance web sight, which has a number of articles on Itanium solutions scaling Linux from 512 cpus and more. The SGI Altix 4700 for example scales up to 1,024 Dual-Core Itanium and 128 TB of globally shared memory: and runs a single Linux system image.
The money is in embedded CPUs
Itanium will never be an embedded processor. It's also works better as a room heater than a number cruncher last time I looked. The latter makes it out of favour with power-budget concious server room admins.
HP pretty much cut its own throat when it discontinued a lot of the old Compaq/DEC stuff such as Tru64 and many customers simply jumped ship rather than moving "onwards and upwards" to PH-UX
Embedded CPUs will never be sexy (except to a very small hard core of geekdom), but they make up the lion's share of unit volume and are cheap to make. Margins per unit aren't high but the sales channels more than make up for it.
WRT comments about multicore linux, given which companies are working on NUMA I'd say a lot less than 5 years is needed and the real driver is the availability of commodity hardware (x86) platforms with that many cores on board.
Can't see it myself. Those guys need the overhead of x86, with its attendant baggage of inherent back-compatibility with the Ark, like they need a hole in the head.
Couple with that that x86 comes in two flavours, crippled and underpowered or capable and speedy but hot as the hinges of hell and I can't see this one going anywhere. Except possibly in the Xbox world, where the market expects a console to cook itself to death in short order and already has a handy acronym for this behaviour.
"the inevitable migration away from costly mainframes"
I do love the way that keeps getting trotted out when talking about how wonderful a sodding great lump of Itanium or similar Big Iron is. The mainframe is alive and well, but relying on VMware and SANs not IBM.
progress, or the way of the world
HP is hardware-free, they make more money out of services (we run your printing shop) and consulting (and sell advice when things don't do what customers want). Why bother with chips, or engineering? Banking beats manufacturing was equally a motto for General Motors.
The real worry is when Intel move to the same idea, and the last chip is eaten by the grizzly.
"Banking beats manufacturing"
"Banking beats manufacturing was equally a motto for General Motors."
I don't know if anybody's mentioned it to you, but half the UK/US bancasinos collapsed last year and ordinary decent working folk had to pledge most of their income tax for the next 20 years to ensure that the City spivs and wideboys continued in the state of luxury to which they'd grown accustomed.
Once that's sunk in, you might want to see what GM's finest have managed to achieve for the company and its employees. (Summary: they're bust).
Actually, the IBM mainframe business seems to be thriving again.
They're the "green" solution now, running thousands upon thousands of linux server images on one ten-ton monstrosity.
Let's try to understand the processor market a bit
 netbooks, cellphones, some gaming consoles and the embedded world. Atoms, ARM, some non-power6-type-PowerPC belongs here. The design and the designers for this team live in a different world from the "big iron" world. The "big iron" world may get wiped off without impacting this market and vice versa.
 The adjacent to the embedded market is the PC & the laptop market market. Again, a very specialized set of designers sit here churning out core2 duos. And again, the entire "big iron" market may get wiped off without impacting this segment that much. Currently, this market is in a strong tussle with the embedded market.. (case in point: netbooks eating into the laptop space).
 commodity server market. This is x86 again. Opterons and Xeons. No core2, ARM, or ATOMs exist here beyond some experimental machines. If this market was wiped off, it would be replaced by the big iron and not the processors from the PC/laptop world (however, the opposite is more likely). Nehalem from this segment is pushing the boundaries of the next segment. Supercomputers is a market this segment has already snatched from the "big iron" world. This segment is almost equally divided in windows servers and linux severs running on Xeon/Opteron (x86-64). SPARC has always overlapped between this segment and the next (and hence Sun is paying the price now. Or rather, it's Oracle who would be making any "pay"ments.). Cisco is trying to enter this market segment. Most (not all) blades belong here.
. Finally, True "Big Iron" segment. Erstwhile Alpha, MIPs and currently SPARC, Power6, itanium play here. This is the battleground for big bad RISCs. The problem is RISCs are selling less and less every year. So yes, this is a slowly declining market segment (shrunk rapidly this year coz of recession; but it may recover some of the lost ground in coming months) . This market is under severe threat from the previous segment (xeon+linux type machines). The x86 world is eating into it slowly but steadily. The main players (ignoring SPARC for now) are Itanium and Power. Both run several OSes - including mainframes and NonStop from HP. The R&D dollars needed to keep these beasts alive are available for now. So the reports of itanium/SPARC/power demise are greatly exaggerated. Yes, in a slightly distant future probably only one of them would survive - but get this - being the last RISC processor standing is not necessarily a good thing for the parent company- coz while you were focused on winning this battle (for crumbs by the time it's over) your competitor wiped off the x86 (server) market.
Sorry about the long post; but I saw just too many out of place comments. Don't blame them; the article itself was talking about gaming consoles in an itanium article. In the end some more points that don't get covered in above explicitly:
- Even if PowerPC running gaming consoles and other embedded stuff wipes out the embedded market - it would do no good to power6 health- which depends on Unix/mainframes business only.
- Mainframes running multiple Linux images doesn't make it "green" - the underlying processor is a power hungry Power6, regardless of what IBM sponsored results tell you. The perf/watt of the processor matters - not how many images of OS it runs. (I do concede that current x86's perf/watt is not any better than Power6 but there are greener options than either to run virtual linux images).
- Mainframes (and nonstop) sub-market is one which linux/x86 can't eat. The machines are based on a different design paradigm and the fault tolerance, High Availability, disaster recovery offered by these systems can't be achieved in a meaningful way (ie in datacenters, not labs) by linux/windows/Unix. So even if power and itanium evolution stops; these baddies will continue to sell for another 50 years.
- While Linux on itanium has scaled to a single image 512 processors in a SGI lab; it'll be a long, long while(even a decade) before customers start using linux on x86 beyond 4/8 processors. CIOs like to stay one step behind. And they matter, not a lab result.
Paris, coz her love doesn't sink; you do.
RE:Let's try to understand the processor market a bit
Everyone whould read Al's post twice. It is so silly for pople to try to compare embedded chips to x86 and x86 to the heavy lifting CPUs of the big iron machines. There are so many business requirments that can only be satisfied on big iron RISC and mainframes, that suggesting that thee platforms will disappear in favor or this or that new x86 chip runing one of the Linux iterations evidences ignorance of large-scale computing and data operations. Some of us live in worlds that many have never dreamed of - where 128, 256 or 512 cores in a single server instance running at 100% continuously for years is SOP. Try to move this to an x86 box running Linux? Now that's funny. We snigger as these knowldge-limited comments.
Sorry Twatt Bryant
This just in, from LJE himself:
"Alright, Oracle’s done integrated hardware and software
design with the Exadata database machine. But Exadata
uses standard Intel chips. Are you going to discontinue
the SPARC chip?
No. Once we own Sun we’re going to increase the investment
in SPARC. We think designing our own chips is very, very
important. Even Apple is designing its own chips these days.
Right now, SPARC chips do some things better than Intel chips
and vice-versa. For example, SPARC is much more energy
efficient than Intel while delivering the same performance on a
per socket basis. This is not just a green issue, it’s an economic
issue. Today, database centers are paying as much for electricity
to run their computers as they pay to buy their computers.
SPARC machines are much less expensive to run than Intel
What makes an Opteron not suitable for a mainframe?
Interesting post from Al.
"The machines are based on a different design paradigm and the fault tolerance, High Availability, disaster recovery offered by these systems can't be achieved in a meaningful way (ie in datacenters, not labs) by linux/windows/Unix."
I happily accept that Linux/Windows/Unix aren't (yet?) real OSes in the mainframe/HA/disaster-tolerant sense, but I'd be interested to hear what's missing from Opteron hardware that stops the top-end software (be it NonStop, VMS, whatever) being implemented on AMD64 rather than on Itanium. I've not heard a good answer yet, and I'm familiar enough with chip architectures to understand a reasonable amount of detail about AMD64 hardware, from an OS implementer's point of view. Certainly I'm not convinced by the unsupported by hard evidence "AMD64 doesn't have the necessary RAS features" waffle I've regularly heard from the Itanium camp.
In replying, please no BS about chip level lockstep. Tandem hasn't used lockstep at processor chip level for a very very long time - see eg this Tandem technical article from 1990:
Re: What makes an Opteron not suitable for a mainframe?
I do not work at nonstop/mainframe (or even Itanium) shops to know what the real reasons are! But here are my guesses (probably incorrect/incomplete) on why IBM/HP are not choosing Opteron(or Xeon) for mainframe/NonStop now:
- IBM/HP have just (just = "in last 5 years") moved their mainframe/NonStop customers from proprietary processors to Power5/Itanium. I am sure this would have required a considerable effort for all ISVs and most customers. And I'm sure both IBM/HP are still spending a few extra bucks to be able to support the remaining proprietary install base along with the new itanium/power installations.
- So moving mainframe/NonStop to Opteron/Xeon would mean - doing the whole exercise again! I am not sure how many ISVs and customers would agree to this. (Imagine a customer from 1970s who agreeed & just demolished the proprietary data center to move to the new machines and payed a multi-million dollar bill. Why would they want to move to Opteron/Xeon in near future? In fact, they would be very, very upset if there is any talk of move away from Itanium/power).
- Supporting 3 processors (instead of currently 2) would again mean some cost to IBM/HP as well as the initial cost of porting. What do they stand to gain at this point ? (One may question that they already sell opteron servers, so where's the extra cost? It's in maintaining multiple versions of the NonStop OS or Z/OS as well as some operational costs at customer sites).
- The mainframe/NonStop machines are multi-million dollar beasts. Saving a few thou on CPUs by moving to opteron is not going to add any significant sweetness to the prices offered.
So there seem to be a few good business reasons which would make it difficult for IBM/HP to consider x86-64 for NonStop/mainframe at this point.
There could be some technical reasons as well :
- This is a very narrow market - A business uses these machines when the continous availability (uptime) is paramount to the business (eg: stock exchange for trading related computations). OTOH, Opteron/Xeon play for the mass market server loads (eg: webservers). Some of the design choices they might have made might not be perfectly in sync with mainframe/NonStop needs, nor would AMD be accomodating to such requests if there was such a dialogue. So while it was easy for itanium designers to consider and implement the RAS features (see: http://labs.hoffmanlabs.com/node/95 for points & counterpoints) - It took a while for Xeon to incorporate them (I guess it may still lag behind itanium on some features).
- Are processor level RAS features necessary ? Maybe some of these features cover one in a billion kind of error. But that's the kind of reliability customers in this segment would like to have. It's like home insurance, if your entire wealth+source of income is your home, you might buy it? (say you are a ho "working from home").
- I don't know for sure, but I thought NonStop still uses Lockstep on itanium platforms (that's why they have two physical CPUs). However lockstep doesn't sound like a technology that can't be implemented on Xeon/Opteron - the primary mechanisms of the Lockstep should sit outside the processor and it shouldn't matter whether it's x86 or itanium. I could be totally wrong here - so do your own research before making any proposals to your boss :) .
- Perfomancewise, Opteron and Xeon's latest editions would have to prove that they are siginificantly (say multifold) better than the latest itanium/power before HP/IBM consider x86 for this market. This would happen only if intel/IBM pull the plug on R&D for itanium/power - which would take sometime.
So all things point to a minimum of 5 (more likely 10 or even 15) years before we see any serious talks of NonStop/mainframe on x86 (if at all). HP and IBM do not do their business to prove/disprove a technical point. They make they investments when they see some potential return. At this point (May 2009) I don't see any additional returns by moving HP NonStop or IBM mainframes to Opteron.
PS: People who want to read an informative account of Itanium saga may try this :
(No, I am not BIll Todd under a pseudonym).
Thank you Al, your time and trouble is appreciated.
Let's cut IBM out of this discussion, for two reasons: As well as the reasons you mention, Power is sufficiently diverse and technically/commercially successful in a niche kind of way that it's not going away in the foreseeable future. Itanium is even more niche, and its niche isn't getting bigger. The question is, is the Itanium niche so small as to put significant numbers of folks off buying Itanium, given the inconvenience you point out of yet another port when (not if) Itanium goes EOL.
Hopefully your words will be appreciated by others, but they're not really news to me (though I'd not seen Bill's little discussion in comp.arch, but similar things have been said elsewhere).
Wrt lockstep: you mention Hoff's RAS stuff. I've seen that too, though it notably omits an Opteron-specific feature comparison. Anyway if you have another look at node95 you'll see that it explicitly mentions that there is *NO* instruction level lockstep on Itanium, which is consistent with seemingly definitive stuff I've seen elsewhere (eg the 1990 Tandem article I linked to), whilst inconsistent with earlier Intel marketing pronouncements at IDF and elsewhere (hmmm).
In fact it's hard to see how Itanium, which is basically a massive cache chip with a weird processor attached, could ever do lockstep sensibly. Random soft errors in caches are inevitable, but how would two separate chips cope with random errors at different times while staying in lockstep?
Unlike what Hoff seems to have said originally in node95 back in 2007, Xeons don't do chip level lockstep either as far as I can tell, although some Xeon chipsets can run two independent memory buses either in "mirrored" mode or as one logical double-width bus, which is sometimes referred to (misleadingly?) as lockstep mode, which Hoff notes in an April 2009 comment. It's not a relevant differentiator between Itanium and A.N.Other anyway, the Nonstop stuff hasn't used chip level lockstep for years.
You'll also have seen Hoff's node95 reference to Unisys, whose "cellular mainframe" used to come in either Itanium or Xeon versions, both of which had extensive in-service RAS analysis features. Unisys have dropped the Itanium one because (they say) they see no RAS advantages with Itanium, and lots of other advantages with Xeon.
"Intel make they investments when they see some potential return."
Indeed. Itanium was justified on the basis that it was going to be the "industry standard 64bit" chip, and the performance leader, and economies of scale would make the price right for such a complex chip. It is not the industry standard 64bit chip, it is not most people's performance leader, and the economies of scale aren't happening, which means that in addition to the production costs, the massive development costs are spread over a relatively small sales volume (which is the reason usually quoted for Alpha's demise). So how long will Intel continue with Itanium, when they would get better return on investment if they spent the money on Xeon, or WiMax, or whatever?
Ah well. Maybe one day I'll get the little details as well as the big picture.
Re: Much appreciated . My pleasure.
This seems to be the first "real" discussion I am having on theregister (I've passively enjoyed many others though). So allow me to continue rambling for one last time.
There really are 2 separate issues being discussed here:
1) Why not Opteron (or Xeon) ?
- That's what I tried to answer last time. The answer is "not now, given the current status".
2) Why Itanium for HP NonStop ?
- That's a question which WAS asked by HP (or maybe Compaq) 6-7 years back. At that time HP/Compaq was willing to bet all it's horses on Itanium. At that time it was imperative to port as many OSes on Itanium - for those who looked at Itanium as a future processor.
- If this question was to be asked in present tense, it's possible that HP's answer may differ. But we are talking of a hypothetical situation now.
2b) If Unisys can move out of Itanium, why not HP ?
- Unisys was never really "into" Itanium, as much as HP. It was riding two boats - Xeon and Itanium - I'm sure an expensive ride - for sometime before they decided to move out completely. At some point they had to choose the one. They made the safer choice - it wasn't very difficult for them to make that choice(Xeon). And it's definitely not a technology based decision, though they'll try to justify it technically to their customers. Whereas, HP was fully into Itanium for all it's bigiron offerings. HP (and IBM) chose x86 for the reborn x86 market. [yup, x86 server market was killed by intel (to push itanium) and resurrected by AMD(to push opteron) - all this happened very quickly].
2b-part2) Why Unisys didn't bet everything on Itanium (unlike HP) in the beginning ?
- Can't say. Maybe they were smarter. Maybe they were risk averse. Maybe the time will tell. (Please note: Unisys is no longer a force in the "big iron" world it used to be.)
Lockstep is called LSU in itanium based NonStop. It is worried about only the reads/writes to memory (ensures they are in sync). See: www.oztug.org/events/2006/AdvancedArchitecture_Massey.pdf
To end, there is one point where I do differ from your viewpoint- Is the Power niche so much bigger than Itanium niche ? According to IDC numbers, Itanium is 66% of Power market as of 2008 (up from 33% in 2007). Smaller, yes. But still the niches aren't that different in size for the business logic to be different for IBM and HP. (PowerPC development has little to do with Power6 development).
I'll get my coat, mine is the one with Paris' pic.
that's not rambling
I'd trade one of yours for any number of ted dziuba spacefillers.
That's a nice presentation from oztug. I interpreted it slightly differently than you though, I thought the LSU was only for ServerNet stuff (ie disk etc but not memory, see the backup slides which talk about the "memory reintegration bus"). Hoff has the same interpretation as you though (node 469 from August 2007), so I'm in the minority, which in a Tandem environment automatically means I must be wrong.
Then again, I found this quote in a really quite nice article from August 2008 in an excellent but little-known journal : "Under HP, the NonStop logical processors evolved to use dual lock-stepped Itanium microprocessors. Because these microprocessors were not deterministic, memory lockstepping could no longer be used. Therefore, the microprocessors were lock-stepped at the I/O level (any packet delivered to the interconnecting ServerNet fabric)."
Does that not make it two each, ie a draw? Now what? Pistols at dawn?
Mine is the one with the Dummies Guide to OpenVMS and Windows NT Integration in the pocket. Who's Paris?
"so I'm in the minority, which in a Tandem environment automatically means I must be wrong."
I do think your interpretation is correct after glancing thru the backup sldies.
Paris is icon of the contemporary business thought. Her purpose is to illustrate how tons of hype creates tons of dollars.
Re: Let's try to understand the processor market a bit
--" While Linux on itanium has scaled to a single image 512 processors in a SGI lab; it'll be a long, long while(even a decade) before customers start using linux on x86 beyond 4/8 processors. CIOs like to stay one step behind. And they matter, not a lab result "--
Perhaps you should take a trip to visit the Leibniz Computing Centre in Munich. LRZ houses Germany's National Supercomputer System, and runs an Altix 4700 with 4,096 Itanium 2 processors, 17TB of global shared memory on a single Linux instance. That got installed in 2006. NASA also have Altrix systems running Linux with 2K+ Itanium 2 chips in them. Big Linux left the lab and proved itself a long time ago.
Re: Let's try to understand the processor market a bit
> leibniz Computing Centre in Munich.
> NASA also have Altrix systems running Linux with 2K+ Itanium 2 chips in them
Real "scale up" market's acid test is Finance (bank of america), retail(walmart), telco (verizon) and manufacturing (Unilever) segments. Not research or academia.
I doubt if NASA or LRZ even have CIOs... probably it's a professor/scientist making the decision (not a CIO like in walmart's case). But don't get me wrong - I am not criticising Linux's capabilities. All I am saying is that "real datacenters" take time before adapting a new technology. For NASA the user is more educated and hands on. They also have a bigger risk appetite than the real world (both due to self confidence and lack of business pressure). For the real world CIOs - NASA, universities, PARC, CERN - it all sounds like "lab experiment" to their ears. (not yours, I understand that part).
And yes, overtime these instances will prove Linux's ability as a single instance champion. That's when HPs and DELLs of the world will start pushing big iron linux as a readily available solution. It will take a good amount of time before we see Linux kicking out Solaris, HP-UX and AIX from the big iron niche.
Oops I think I got my coat aleady.