Sun's president and chief executive officer Jonathan Schwartz declared yesterday that the company's long-planned Rock processors remain on track for release this year. This is good news for Sun watchers. With two rounds of layoffs that server and operating system maker Sun Microsystems announced in 2008 biting into the company's …
Some fellow decided years ago that floating point performance was not important anymore for SPARC because all that was important was database and therefore integer performance. I wonder if he could be found and brought forward to explain himself now?
Rock is dead...
Matt Bryant, HP Analyst, where are you?
Most processors from the 386 on have easily handled that many threads (and many more).
What I think the author means when he says "thread" is actually "process", i.e. an independant execution unit
Rock Systems In OpenSolaris
4.1.1. Supernova platform names
Supernova platforms will have the following root node "name" and
Contents: Root node name property for Supernova platforms
Value: "SUNW,SPARC-Enterprise-AT7480" for Silver-II
Value" "SUNW,SPARC-Enterprise-AT7880" for Platinum-II
Contents: Banner name for Supernova platforms
Value: "SPARC Enterprise AT7480" for Silver-II
Value: "SPARC Enterprise AT7880" for Platinum-II
I believe the AT7480 will be 4 Socket and and AT7880 will be 8 Socket.
When pigs fly
'09 is gonna be a downer worldwide economically.
no, the article is correct when referring to 256 threads... as it doesn't refer to OS threads, but CPU threads... it's like you had 256 independent CPUs in one
Matt Bryant and Co
Gonna be a sweet, sweet thing watching lil' Matty-boy swallow the screeds and screeds of bullkaka him and his oddfellow "Rock is Dead" pal have served up on the high-profile industry news hotwire that is the comments section of El Reg for the best part of two years.
Here Matty Matty Matty
Hardware threads are essentially virtual CPUs which allows the operating system to despatch several independent execution threads at the same time. However, several threads share the resources of one core (from 2 hardware threads on a hyperthreaded Pentium up 8 on a Niagara T2 processor). The core itself can have more than one set of processing resources (the Niagara T2 has 2 integer and one floating processor per core shared over 8 hardware threads). Threading also varies in its sophistication - the Niagara has it at the very basis of its being, on SPARC64 it's a retro-fit.
All these techniques are essentially there to make use of core resources which would be otherwise idle whilst wating to be fed with data or instructions (a modern CPU can execute a hundred or more instructions in the time taken to fetch data from main memory in the event of a cache miss). The downside is that, whilst total throughput can go up, single thread speed suffers when hardware threads on the same core start contending.
The best way to think of hardware threads is a virtual CPUs implemented in hardware.
I know a number of enterprises who are waiting for the Rock. It could be such a massive game-changer they are happy to wait for it. Of course, if its a massive flop then they're just buy the SPARC systems they were going to buy anyway.
Sounds like Sun are stuck between a Rock and a hard place ...
Re: Here Matty Matty Matty
You want to watch that - he'll have your IP address, track you down, threaten to drown kittens etc.
OTOH, you can't spell 'analyst' without 'anal' ....
New Product into a declining space
Sun's between a Rock and a hard place.
Losing Unix market share to HP, IBM and x86 industry standard servers.
Unix is just too expensive for the majority of applications and Sun's going to have a difficult time basing their recovery on this product in the worst global economic environment for decades.
Don't bet on the 3 legged horse
After over a year of screaming at sun about a dodgy T1000, we get a brand new shinny replacement. While it was much faster than the lemon they 1st sent, its a dog running anything but the most contrived code. Compared to a Netra 210 which costs the same and uses a old SPARC IIIi, the T1000 is about 1/2 the speed at best if given a highly parallel task that was poorly written. Give it optimized code and it the T1 cpu just chokes. It is only good at moving lots of memory around or endless chasing of pointers, both which are signs that the programmer should have been sacked. I now have less than 6 months to order the last of the small 1RU servers that still run Solaris 9 which is a shame since they are Rock Solid unlike anything I've seen out of Sun lately. I've been told Sol 10 will solve all my problems but I don't see needing reboots after monthly patches to fix security bugs as a feature. I sort of like to have machines with about 4 to 12 packages installed and not needing any reboots for years as good thing but apparently that isn't progress and I need to step up to the Microsoft way and embrace the new features of Solaris 10 with its requirements of heaps of packages including the kitchen sink packages.
Lame answer from Pony boy
"I think we've said to plan on it for later this year, and we are still on track," Schwartz said.
too bad he could not say anything positive like:
- The wattage is no longer 500Watts
- The transactional memory finally works
- We have agreement from Oracle that they wont change 12 licenses per chip
Basically he said.... "I will not admit that we plan to RIF ROCK"
Re: Lame answer from Pony boy
Since you reference www.sunsucks.com (a HP run site mind you), I must believe
that you think HP has the answers... Let's see:
1. ROCK is late, but Itanium is always late. Have they had on time delivery yet? Niagara was on time. Where's tukwila? Or will Tukwila even be released?
2. 500W is typical HP FUD. The numbers I've heard of aren't even close to that. See:
Even at 230Watts that's only 230/16=~15 watts per core. That will blow away Itanium.
3. Transactional memory finally works? What? It hasn't even been released and you pull that out of Matt Bryant's butt? Gimme a break. Who knows if it will work? That's for the theoreticians
to figure out. It has nothing to do with anything, however, as Intel nor IBM have anything like it, so would you rather have something that might help but doesn't hurt, or not have it at all? Pure FUD.
A question for you HP boy. Since HP has outsource all of their family jewels... what will you do when Intel junks Itanium? You'll only have three choices... IBM/X64/SPARC.
All in all, ROCK looks very interesting and no one else has anything even close. Also, since SPARC is open, even if Sun screws up, someone else (Fujitsu) can come along and help out. Pretty nice protection for SPARC users. HP has no such protection as they have to rely on another company with completely different goals for everything as they have no real IP.
@Tim H and Pony Bot
Tim ... we're happy with the Niagara stuff here but a decent legacy solution for Solaris 8 and 9 would be handy - the whole Brandz and container\zone thing is just a PITA. Got a load of V-series kit in store so it's not too bad for us.
AC\Pony Boy - WTF? We have a lot of HP-UX as well and the future looks a damn sight less clever for Itanium and PA-RISC. HP Seem to be betting on Linux so they can shift BL blades but, frankly, we can get x86 stuff from anyone we choose.
Re: Lame answer from Pony boy
Hm... that kind of mentality is why we're stuck using sucky x86 processors when we should all be on 64-bit PPC's.
- HP killed its PA-RISC arch, and switched over to Intel trash. Farewell to the HP9000's, the only good thing they had left. HP-UX on Intel isn't just as good anymore.
- IBM's basically juicing up its Power chips, they're actually doing improvements on their chips. Oh, and their Cell BE chip is also an innovation with its SPE cores.
- Mac... doesn't even sell servers (?) and also went to the Intel arch. Losers.
Sun is actually innovating, and making a bold commitment with its Rock chip, which deserves a lot of respect in my eyes. Their current SPARC servers are also in high regards, as I've seen a lot of x86 chicken-boxes fall down under high loads, but the almighty E25K still stood rock solid under similar loads.
I really, really hope Sun succeeds with this project; its about time we had something really new and not yet-another-rehashed x86 arch.
Oh, and Sun... bring us back the SparcStations!!!
Dear Sunshiners..... (and Bill)
Apologies, I've been busy with some new storage software. I'll give you a clue - it can manage Sun arrays but it's not written by Sun because it can also manager and monitor just about all the other major player's storage kit too (and can even run on Red Hat).... Oh, sorry, that's far too confusing for you Sunshiners, I forgot that Sun doesn't do storage management software. Tell you what - wait until the other Sunshiners aren't looking, then type www.hp.com/go/storageessentials into your browser....
Anyway, where were we? Oh, yes! The return of the Rock Vapourware™.
"I think we've said to plan on it for later this year, and we are still on track," Schwartz said. Supposedly the most important chip design in Sun's future, the one which Sun have banked their future on, and Ponytail "thinks" he told us to plan for it this year? He can't even remember for sure what Sun have told their customers! Mind you, Rock has jumped the tracks so many times who knows where it's at, and definately not Ponytail. With that kind of inablility to reassure board-level decision-makers (they kinda like people to KNOW about their products, not THINK they know), is it any surprise they're deserting Sun's high-end products in droves?
So Ponytail has to stand up there and fan the flames of belief, as unless he says something positive the press will be screaming "Rock is dead" in a flash. Ponytail needed to stand up there and say the problems with scout threads and transactional memory are fixed, the design has gone gold, and everything is groovy, baby. So the best he can do? He "thinks" he told us. Inspiring - not! But then it's also revealing, the lack of definity in not setting a date for Rock is fatal. By not putting a stake in the ground he instead puts a stake through Rock's heart. Now everyome is wondering "If he only thinks he told us, does that mean he isn't tracking the project closely, maybe the bugs aren't fixed, maybe this year could become end of year or even next year?"
The doubt is obviously infectious seeing as how TPM keeps using the "if" word whenever he refers to Rock in the rest of the article.
"....A question for you HP boy. Since HP has outsource all of their family jewels... what will you do when Intel junks Itanium? You'll only have three choices... IBM/X64/SPARC...." Well, seeing as both hp and Intel make money from Intanium (more than Sun is making from SPARC by a country mile), and both hp and Intel have plenty of cash for develpment, marketing, an increasing market-share and do not have poor market ratings like Sun, they therefore have no reason to kill Itanium. But then, hp also have a diverse portfolio of products (did I mention their storage software?), which measn they have other healthy sources of revenue just in-case the extremely unlikely event of Itanium being SPARCed happens (about the only chance of that happening would seem to be if Ponytail got the CEO job at Intel!).
"....All in all, ROCK looks very interesting and no one else has anything even close...." True, the other vendors have actual chips whereas Sun have a vapourware design that doesn't work in the labs and doesn't even have a solid release date.
"....Also, since SPARC is open, even if Sun screws up, someone else (Fujitsu) can come along and help out. Pretty nice protection for SPARC users...." Really? So you didn't notice that big thing of FSC splitting up, which should keep Fujitsu busy enough managing along with the falling revenues and laying off workers in Germany (hard to do so must be bad). You also failed to notice Fujitsu is still signed up for the Itanium Alliance, but has not renewed the SPARC agreement with Sun, and has not announced any intention to make Rock-based servers yet, instead only announcing SPARC64 VII+ on a rather sparse roadmap. All in that doesn't sound very reassuring for SPARC users at all! But don't worry too much, there's this really fun software from Transitive that lets you run SPARC binaries unchanged on x64 and Itanium, just to tide you over until you get round to migrating to Power or Itanium.
@ Matt Bryant
X86 under load...
How many chickenboxes with a load balancer and fast interconnect for the price of a ripoff box? Obviously YMMV but redundancy and scalability is cheap today for many applications. No wonder blades sell.
isn't the problem that it's gonna be a little to little and much to late.
I mean if SUN get rock based systems out in 2H, about the same time as Tukwila, and then AFAIK POWER7 will be just around the corner, so IBM will most likely already have benchmark results out.
And that will mean a hard time for a system that was originally slated to compete with power5+ and Montecito based systems.
And still they will need single threaded performance, this is the main issue that I hear/encounter as being a problem, with the current Niagara based servers.
Just a while ago I heard the lead SAP architect where I work, complain that it was the second time that he had been involved in migrating a customer from an oldish SUN server to a new niagara and that the throughput had sucked. The new system actually ran slower than the old one.
So yes it is good for SUN that if they get Rock out of the door in 2H, buuuutt.. still it's late to the game.
@ Matt Bryant
"Well, seeing as both hp and Intel make money from Intanium (more than Sun is making from SPARC by a country mile), and both hp and Intel have plenty of cash for develpment,"
Based on the current climate where no company or product is safe, I'm willing to bet Itanic is at the top of the scrapheap for junking by Intel as they cut jobs, close plants to try and cut costs like everyone is doing right now. Itanic is dead, no one cares about it.
All systems above 8 sockets cancelled for ROCK
ROCK 4 core (16 mini-cores)
small cache per core and only 8 sockets max.
Sounds like it could be a great system in early 2007 which was the originial date.
Keep taking the meds, moonshiner Matt
"....Based on the current climate where no company or product is safe...." Blimey, is a Sunshiner actually admitting things aren't rosey and that no product is safe, even Sun's!?!? Oh, sorry, no he was just implying the whole recession thing affects everyone BUT Sun. Of course, the fact that Sun is in such a much worse position than Intel or hp escapes him, otherwise he'd be forced to admit by his own logic that Rock's future is massively more perilous than any Intel chip.
"....I'm willing to bet Itanic is at the top of the scrapheap for junking by Intel as they cut jobs, close plants to try and cut costs like everyone is doing right now....." Well if it was so top of the heap then why did they close plants, lay off workers and cut costs then? Why didn't Intel just kill Itanium if Intel's got it so bad? Or could it be the layoffs were in x64 related areas because that is the line that has already seen and probably will see the largest drop this year. And could it be that both Intel and hp are making money on Itanium just fine, especially in the high-end space where x64 doesn't fit, so it actually makes strategic sense for both hp and Intel to keep the Itanium train rolling. Oh, of course not, silly me, we're using Sunshiner Logic™ now, where the worst things will always happen to Sun's competition just becasue Sunshiners want it to.
Don't get me wrong, I think anyone looking at Fujitsu's SPARC64 VII+ kicker as an upgrade will be fine, much more likley to be both on-time and a good performer than Rock, but Fujitsu haven't exactly waxed lyrical over what comes after that. After that, if you want to use Slowaris in a high-end system then you better hope IBM don't mess up on Transitive.
Re: All systems above 8 sockets cancelled for ROCK
I don't know about canceled, but 8*16 = 128 cores * 2 Threads = 256 Threads in only 8 sockets of space. Also, I hear that it will come out at 2.3GHz, which is better than the 2.0 GHz that Tukwila is rumored to max out at. The power requirements must be low when compared to comparable systems as well.
Re: Fujitsu haven't exactly waxed lyrical
Well, Matt, what I do know is that Fujitsu, as most vendors now, have lost faith is Itanium. Intel gives vague rumors of a possible successor to Tukwila, but seeing as Tukwila is almost two years late, I would bet that they are hedging their bets on that one. Fujitsu see's the writing on the wall and are pushing SPARC much harder than Itanium. SPARC has a much brighter future than Itanium as there's actually more than one vendor using it.
Also, seeing as SPARC is open, I wouldn't doubt that there is another manufacturer out there seriously thinking about coming out with a SPARC chip of their own. So much for the closed Itanium "standard". Thank god that AMD saved Intel's ass by coming out with 64-bit extensions to x86. Good think Intel was smart enough to pick up on the 64-bit extensions. Itanium's a mess.
Re: RE: deadbeef
"Don't get me wrong, I think anyone looking at Fujitsu's SPARC64 VII+ kicker as an upgrade will be fine, much more likley to be both on-time and a good performer than Rock, but Fujitsu haven't exactly waxed lyrical over what comes after that. After that, if you want to use Slowaris in a high-end system then you better hope IBM don't mess up on Transitive."
That's funny Matt, I heard from Fujitsu and Sun themselves the following:
"SPARC64 VIII will have eight cores, be manufactured in a 45 nm process, and run at 128 GFLOPS per socket."
What're Intel's plans?
Sounds like you aren't as privy to as much info as you think Matt.
RE: Re: All systems above 8 sockets cancelled for ROCK
"I don't know about canceled, but 8*16 = 128 cores * 2 Threads = 256 Threads in only 8 sockets of space...." Which also implies 128 wheiner cores without the space for enough memory to keep them ticking over. It's just Niagara only writ bigger and even more expensively. Compare to IBM's pSeries or hp's Integrity, where the memory available is large enough to keep the man-sized cores ticking over nicely. Then consider that both pSeries and Integrity will scale to much higher socket counts and you realise that Sun have capitulated on the high-end space.
"....Also, I hear that it will come out at 2.3GHz, which is better than the 2.0 GHz that Tukwila is rumored to max out at...." Oh dear, you still think the Megahertz War is still on? I have an old 3GHz P4 somewhere, do you think that will out-perform Tukzilla just because it's got a faster clock? And that's ignoring the fact the number most quoted seems to be a 2.8GHz starter for Tukzilla. Instead, try thinking about what the core is actually doing each clock. Tukzilla's register count and large cache will make it much more effective with the kind of heavy-thread apps customers in the high-end actually have today, rather than the nonexistent massively-parallel apps Ponytail keeps promising them will come. Face it - even if Rock ever gets sold it will have a hard time competing with today's Xeon, let alone tomorrow's.
Industry need Innovation
Sun needs Rock to be a success but so to does the rest of the industry.
Innovative architectual improvements are required to further increase
performance. Just putting more cores on the chip is a dead end. If
hardware scouting and transactional memory provide Rock with a worthwhile
performance improvement then the other vendors will be encouraged to
investigate similar improvements in their CPUs.
After Sun I'd rate IBM the most likely to try something new.
Its worth remembering that events often do not play out as planned. The
Itanium market certainly has not evolved as Intel would like it to have.
Without the success of the AMD Opteron Intel would not have been compelled
to extend the x86 architecture to 64 bits and so create its own competitor
to Itanium. A funny HP story illustrates this point.
A few years ago I was at a HP presentation on new products. The presenter
admitted that demand for Opteron based servers had been greater then
expected. They then went on to explain that the Opteron servers were really
only 32 bit machines with 64 bit extensions while the Itanium based
Integrity servers were true 64 bit machines. I could see that the presenter
absolutely did not want the largely non technical audience to get the idea
that the cheaper Opteron machines could in anyway be a viable alternative to
the more expensive Itanium machines.
I decided not to point out that the argument by which the Opteron servers
weren't truly 64 bit machines was equally applicable to HP's own 64 PA-RISC
machines which were also originally 32 bit. Surely not a argument that
would be appreciated by HP's marketing department.
Never believe any SAP specialists ... SAP Software sucks, uses GB's of RAM doing nothing, poorly designed and heavy backend, crappy ui.... basically, SAP NetWeaver is such a big piece of cr*p.
We Germans call it NetWieder, which means as much as "not again". In fact, NetWieder is much better than the previous sh*te they brought. All the "techy" needed was an excuse to explain why SAP runs slooow.
Best Run Business AVOID SAP!
RE: Re: Fujitsu haven't exactly waxed lyrical & Re: RE: deadbeef
"Well, Matt, what I do know is that Fujitsu, as most vendors now, have lost faith is Itanium. Intel gives vague rumors of a possible successor to Tukwila, but seeing as Tukwila is almost two years late, I would bet that they are hedging their bets on that one...." If Fujitsu didn't have faith they would have canned their Itanium range, which they haven't done. Intel have a roadmap for Itanium that stretches further than any SPARC roadmap (I have a version reaching to 2020 but it's NDA so you'll have to go get your own). The SPARC64 roadmap is so vague it could have come out of a Rock planning session, just check back and see how unimpressed The Reg was back in October http://www.theregister.co.uk/2008/10/15/sparc_roadmaps/
"....Fujitsu see's the writing on the wall and are pushing SPARC much harder than Itanium...." Actually, Fujitsu aren't pushing either particularly hard, which is interesting. They long ago realised they can't compete with hp in the Itanium server market so I'm not too surprised they don't scream too much about their Itanium servers, but I haven't heard too much out of them about SPARC64 VII+ either. That's not like the old days when they would take any chance to come and tell us how much better and cheaper SPARC64 was than UltraSPARC (and it was).
".....Also, seeing as SPARC is open, I wouldn't doubt that there is another manufacturer out there seriously thinking about coming out with a SPARC chip of their own....." Who then? What, someone making washingmachine control chips? Or calculators? Because no major chip manufacturer has expressed an interest, you're just regurgitating another Sunshiner pipedrean. Don't tell me, in your fantasy it's Jobs standing up to announce an Apple SPARC! What comedy!
"....Good think Intel was smart enough to pick up on the 64-bit extensions. Itanium's a mess." Well, if you mean a mess is taking high-end marketshare from your competitors then I bet Ponytail wishes he had more messes on his hands, then maybe Sun wouldn't be neck-deep in the brown stuff like it is.
"....That's funny Matt, I heard from Fujitsu and Sun themselves the following:..." And I suspect you heard that from Sun, not Fujitsu, for the reason below.
"...."SPARC64 VIII will have eight cores, be manufactured in a 45 nm process, and run at 128 GFLOPS per socket"....." Yeah, and the date for launch is.... oh, there isn't a date. That's because Fuijtsu wouldn't confirm it and Sun is pressuring them to supply a date becasue they know Rock isn't going to arrive on time if at all. Sun are desperate to have SOMETHING they can announce, but Fujitsu won't be pinned to a date. Don't worry, I get the Fujitsu updates too, and they declined to give us a fixed date for anything after SPARC64 VII+. But you go on believing if it makes it easier for you.
"Yeah, and the date for launch is.... oh, there isn't a date."
...and this is different from Itanium, how? Tukwila is two years late and running. They have no solid dates for any future releases. Tukwila was supposed to have 8-16 cores running at 4.5GHz and be released in 2007. Now it is 4 cores, 2.0GHz and to be released in 2009. At least ROCK has stayed steady and provides something new, whereas Tukwila just increases core count and uses a new PC-based interconnect (with lots of cache).
Floating Point Note for Anonymous Coward
"floating point performance was not important anymore for SPARC because..."
Ummm... The T2 and T2+ processors topped floating point performance from the day they were released until recently!
You should not paint all SPARC processors from a single T1 processor release when the rest have performed very reasonably well or superior to the rest of the market!
Goat Jam: Thread Clarification
"Most processors from the 386 on have easily handled that many threads. What I think the author means when he says 'thread' is actually 'process'"
When you look at traditional processors, every time a new "process" needs to be swapped in, the CPU must push the registers back to memory.
Some CPU Architectures will switch between "threads" of execution without pushing all of the registers out to slow memory and the same memory segments are often leveraged by the thread of execution.
When you look at SPARC processors, registers do not necessarily have to be pushed out to slow cache during a process switch, meaning SPARC scales better with multiple processes than traditional CPU's.
When you look at the T1 & T2 processors, the CPU is not pushing registers out to slow memory, and the hardware threads can operate on different memory segments simultaneously without pushing program execution address registers out to slow memory.
To understand why SPARC and CoolThread T1 & T2 scale to hundreds of simultaneous threads more efficiently than traditional processors and millions of processes effectively when traditional processors will merely thrash with their applications timing out - one must understand architecture.
SPARC allows for much heavier utilization than traditional processors while SPARC CoolThreads allow for massive utilization over traditional processors - all because of threading.
This does not mean that SPARC is perfect for every bill - but one must be educated to understand when it does fit best.
Matt Bryant Forgets: Using HP Storage
"I forgot that Sun doesn't do storage management software"
Storage Management requires software... Open Solaris, ZFS, and GUI's are all software.
Your storage is faulty, Matt - better migrate to ZFS so you don't forget so much!
Matt Bryant: Spelling, Grammar, SUN & Fujitsu Announcements, Intel Plans
"Sun are desperate to have SOMETHING they can announce, but Fujitsu..."
You should have started, "SUN is..."
Announcing something that will happen more than a couple quarters in the future is not very helpful, since it would eat into current system purchase margins.
The only one desperate for SUN or Fujitsu to make an announcement that may be one or more years out are people who are interested in seeing the sales of those companies demise.
It seems Matt Bryant is desperate!
"Intel have a roadmap for Itanium that stretches further than any SPARC roadmap (I have a version reaching to 2020 but it's NDA..."
I remember Intel recently canceling a bunch of silicon lines and moving to multi-core, emulating SUN and their partner companies (Fujitsu & AMD.) Some of those former Intel NDA'ed plans fizzled.
Then again, Matt Bryant could be a liar.
I see an IT jihad!
Reading these comments, what I'm seeing is religious zeal, not reality. I've used Sun gear for 20 years. Generally good stuff. I have some max'ed out E25Ks that need to be replaced. I've been waiting and waiting and waiting for Rock and enduring excuse after excuse from my Sun sales team. So now my CIO and CEO have told me that our waiting is over; we will replace the E25Ks by year's end. What does Sun have to offer? A bunch of little one, two and four socket boxes, running Niagera. For multiple, high transaction, terabyte siszed databases? Yeah, right. The single-treaded performance on Niagera is horrible and I don't have the time or money to rewrite all our software to run well on CMT. CMT is great for application servicing and the like, but it isn't for heavy lifting (though it appears a lot of people have been trying to pound that square peg into a round hole and calling it "innovation"). Of course, Sun has told me that I could buy some M9000 machines to tide me over until Rock is ready, but theuy didn't offer me any of what they have been smoking.
Consider this: If Rock is really on its way for late 2009, is supposed to blow everything else out of the water, is completely binary compatible with UltraSPARC IV+, and will be introduced on High-end Supernova gear, why did sun just extend their OEM agreement with Fujitsu for four years? And why are they still hawking SPARC64 gear? Why does Sun only want to talk to me about Fujitsu gear? That is, when they want to talk to me at all; they've become very, very quiet.
Consider this, too: Sun announced a major reorg in December. The company is being reorganized to focus on the application servicing and the next fad - cloud computing (as if this were something new and exotic). What do you thing their acquisition of Glassfish, MySQL and others is all about? Rock, and high-end gear in general, is inconsistent with this roadmap. It doesn't make any sense.
Even if Sun rolls out Rock on schedule, would anyone in their right mind bet the farm on version one of anything from the choas that is now Sun? They have had severe quality problems over the last two years. They rush products to market without thorough testing. We've had terrible issues with nearly everything Sun has rolled out recently. There's a good reason why Sun has been sliding down and left in Gartner's Magic Quadrant.
Everything Sun has done over the last year has had the purpos of preparing the company for sale. Their market cap has fallen by half since last July, they've got an infusion of investment money, they've laid off nearly 20% of their workforce, and restructured their debt to some degree. Sun is for sale; no doubt about it. I predict that they'll sell off the SPARC business to Fujitsu. Who knows who will buy or merge with them for the rest of their porduct portfolio.
As for HP sunsetting their PA-RISC chip in favor of the Itanium EPIC cips: stupid, stupid, stupid. They are forcing all of their high-end gear customers into a very expensive platform migration. This has IBM grinning like the Cheshire Cat! Sun/Fujitsu, too, to some extent. If one is going to have to migrate their big iron platforms, due diligence dictates that you evaluate the competition. And this is where IBM shines. They've cranked up any number of programs to intice Sun and HP customers; not just marketing hype, but deep discounting and free migration services. They abandoned Itanium and for good reason. It is a niche chip, without a clear roadmap and strong comittment from Intel. Itanium is already old technology and IBM's POWER6 chip, with it's dual cores and 5 GHz clock speed is a generation ahead and IBM is readying the POWER 6+ chip now. They are also on track with POWER7. It won't be late; IBM is never late to market. It is doubtful that Intel is going to continue to dump money into Itanium, when it has so little market share and an almost non-existent adoption rate. Their making money on it, but not enough to justify the investment tens-of-billions of dollars needed to advance the chip beyond the already late Tukwila. It is very likely that Integrigy Superdome customers will be faced with yet another expensive plaform migration in just a few years. Why take the chance? The old saying that no one has ever been fired for buying IBM. For some time now, this has been true of HP as well. But I suspect that there may be quite a few IT managers getting the can for choosing Itanium-based Superdomes over IBM POWER.
Anonymous: SPARC, Power, Itanium & Migration...
"So now my CIO and CEO have told me that our waiting is over; we will replace the E25Ks by year's end. What does Sun have to offer?"
Of Course, the SPARC64!
SUN has always resold other vendor CPU chips in servers, I don't understand why people are so concerned about Fujitsu. CoolThreads was made by another vendor before SUN bought them. There was the "e" and "i" families of the UltraSPARC III processors, which was a different family. TI manufactures chips for SUN. What is the problem here?
"why did sun just extend their OEM agreement with Fujitsu for four years? why are they still hawking SPARC64 gear?"
To provide a long life for the M series of servers from Fujitsu Partnership. SUN had kept the UltraSPARC III, IV, and IV+ around for A LONG TIME - even when SPARC64 was released! Keep in mind, Fujitsu was selling SPARC64 systems before SUN resold them and Fujitsu will most likely continue selling SPARC64 systems after SUN releases new processors. This is the beauty of OpenSystems!
"Why does Sun only want to talk to me about Fujitsu gear?"
It is not Fujitsu gear, it is gear partnered and sold by both companies. Fujitsu resells CoolThreads servers, by the way.
"What do you thing their acquisition of Glassfish, MySQL and others is all about?"
To smooth out the large peaks and valleys in the big iron sales. If you watched the latest quarterly announcement video, you can see how SUN is using software to smooth out quarterly results.
"Rock, and high-end gear in general, is inconsistent with this roadmap. It doesn't make any sense."
Once again, if you watched the quarterly announcement, you will see how hardware sales jump up and down every couple of quarters, and SUN is using the growing software portfolio in order to smooth out the profits & losses. The hardware business provides SUN the lion share of revenue while the software revenue is still only a sizable percentage.
The plan for ROCK is very consistent with SUN's spoken direction for the past decade.
"would anyone in their right mind bet the farm on version one of anything from the choas that is now Sun?"
This is why SUN has multiple lines of systems and Fujitsu will continue to be providing chips for the M series while whoever the chip manufacturer is will start to provide chips for the new systems.
There is more security with multiple vendors (i.e. SUN, Fujitsu, TI, Atmel, Cyprus, Solbourne, ROSS with SPARC) in an open architecture than a single vendor with a proprietary architecture (i.e. DEC with Alpha; HP with PA-RISC; SGI with MIPS; IBM with Power; Intel with itanium.)
"Itanium is already old technology and IBM's POWER6 chip, with it's dual cores and 5 GHz clock speed is a generation ahead"
SUN & Fujitsu (with SPARC) have been at 2, 4, and 8 cores for some time now as other vendors are still playing around with silicon for 4 cores. With multiple SPARC vendors moving to 16 cores on both their low, midrange, and high end lines - I hardly see IBM as a single vendor with as a "generation ahead".
IBM as a single vendor has a faster single thread, SPARC with multiple vendors has more throughput per server, x64 with two vendors has cheaper equipment.
Each architecture has their advantages & disadvantages and SUN plays well in every architecture category now (including IBM POWER, supplying OpenSolaris on POWER, as well as SPARC on IBM Blade chassis.)
It's unlikely but...
Just imagine if Marc Trembley has pulled the rabbit out of the hat with Rock and it really does kick serious ass. And then imagine if the MySQL coders had been able to exploit the transactional memory support. Hmmm, you might be looking at a serious piece of kit. Also, don't forget Sun has a nice relationship with SAP and still talks with Larry Zen Ellison - so there are a number of apps that could be tuned to exploit Rock's new abilities. Who knows - but at Sun's current valuation, how bad can it hurt to speculate with a few $$$?
(Paris coz she knows how to Rock)
RE: Matt Bryant: Spelling, Grammar, SUN & Fujitsu Announcements, Intel Plans:
".....Announcing something that will happen more than a couple quarters in the future is not very helpful, since it would eat into current system purchase margins...." So you missed the bit where Sun have been promising Rock for how many years? Or the bit where they told customers to wait for Rock and not to buy the SPARC64 kit, then had to volte face when Rock got postponed AGAIN, and now are doing the same for SPARCVII+ becasue they think Rock will land this year?
"....The only one desperate for SUN or Fujitsu to make an announcement that may be one or more years out are people who are interested in seeing the sales of those companies demise...." I would suggest there are a large number of Sun shareholders rather keen to know when (or if) Rock or SPARC64VIII will arrive, and I doubt they want Sun's sales demise. And then there are the customers that have grimly stretched their existing SPARC systems in an attempt to tide themselves over until Rock arrives, I'm sure they're quite keen to hear a firm date. Drop the paranoia and look around, one of Sun's big problems is that they have managed to generate a whole cloud of doubt amongst what used to be their customer base, their shareholders, and the market in general.
"...Then again, Matt Bryant could be a liar." And you could be the King of Siam. Simply labelling anyone that disagrees with your point of view a liar is not a very grown-up debating strategy, it just makes you sound like a petulant child.
RE: I see an IT jihad!
Agreed with you on almost every point until you started making out Power was The Only Solution. Whilst it is a much better solution than any current or planned Sun kit, what you negelcted to mention is that Power6 and Power7 are moving in the EPIC direction with features like in-order-execution. You also missed the fact that it is hp Integrity that is eating the biggest share of ex-Sun customers, as shown by the greater growth of hp Integrity in the high-end than IBM's Power. And that is an area where Xeon doesn't go, so it makes strategic sense for Intel to keep reaping in the profits. And a measure of how much a threat IBM view Integrity as is the speed with which they attacked and swallowed mainframe emulator PSI.
RE: David Halko
"....Of Course, the SPARC64!...." Erm, hate to burst your bubble (OK, actually I'm enjoying it), but I sense that SPARC64 is not on the list of choices, but Power probably is.
"...."What do you thing their acquisition of Glassfish, MySQL and others is all about?" To smooth out the large peaks and valleys in the big iron sales...." What, you want a nice and smooth downward arc rather than the current jaggedy-but-relentlessly-downward thread? How neat and tidy. I suggest the real reason is Ponytail's desire to convert Sun into a software only bizz, and the hardware is just being spun out to try and provide the funds until the cloud becomes a reality.
"....The plan for ROCK is very consistent with SUN's spoken direction for the past decade...." Flip-Solaris-flop-Linux-flip-Windows-flop-Solarisx86-flip-SPARC-flop-x64-flip-Niagara-flop-Rock-flip-SPARC64-flop-Rock-flip-SPARC64VII.... You mean Sun had a strategy!?!?! More like Sun had two dozen incompatible and unprofitable strategies.
"....There is more security with multiple vendors (i.e. SUN, Fujitsu, TI, Atmel, Cyprus, Solbourne, ROSS with SPARC) in an open architecture than a single vendor with a proprietary architecture (i.e. DEC with Alpha; HP with PA-RISC; SGI with MIPS; IBM with Power; Intel with itanium.)...." Only if the market believes those multiple vendors will provide a viable and performing product when required with support, upgrades and a mappable future. Please show me the TI, Atmel, Cyprus, Solbourne or ROSS roadmap that has an enterprise SPARC CPU capable of competing with Power and Itanium, let alone Xeon and Opteron. Oh, you can't - what a surprise!
"....SUN & Fujitsu (with SPARC) have been at 2, 4, and 8 cores for some time now as other vendors are still playing around with silicon for 4 cores. With multiple SPARC vendors moving to 16 cores on both their low, midrange, and high end lines - I hardly see IBM as a single vendor with as a "generation ahead"....." Yet again the Sunshiner fails to see the facts - putting the wrong core design in multiples on the same chip still leaves you with a product the customers don't want and the market doesn't have faith in, hence Niagaras inability to climb out of a narrow niche and the questioning of the Rock design, especailly given that it is so late if not stillborn.
Let's give a very simple example from a friend - the only message Sun is actively pushing into a major UK telco (an account that used to be 60% high-end Solaris on SPARC) is why don't they put Slowarisx86 on ProLiants and buy support from Sun. I doubt they are the only ones. SPARC is dead, Slowaris may live a bit longer through OpenSolaris, but I can't see that generating enough revenue for Sun without massive downsizing.
RE: Kevin Hutchinson
"....Just imagine if Marc Trembley has pulled the rabbit out of the hat with Rock and it really does kick serious ass. And then imagine if the MySQL coders had been able to exploit the transactional memory support. Hmmm, you might be looking at a serious piece of kit. Also, don't forget Sun has a nice relationship with SAP and still talks with Larry Zen Ellison - so there are a number of apps that could be tuned to exploit Rock's new abilities. Who knows - but at Sun's current valuation, how bad can it hurt to speculate with a few $$$?..." I think you'll find most investors will want a bit more fact to base their decisions on, rather than wild imaginings. But if you feel that strongly then go ahead and bet your mortgage, it might even pay a few minutes of Ponytail's rediculous annual bonus.
Matt Bryant: Spelling & Debating Strategy
Anonymous: "...Then again, Matt Bryant could be a liar."
Mat Bryant: And you could be the King of Siam. Simply labelling anyone that disagrees with your point of view a liar is not a very grown-up debating strategy
It seems you titled your response as "RE: Sunshiners" - labeling them as such.
Spell correctly and take your own advise.
Itanium sales forecasts ... still haven't got a clue do you MB?
matt - have a look at the following web site to remind yourself about the illustrious Itanium history - http://en.wikipedia.org/wiki/IA-64 - this was Intel's expectations, largely driven by HP? Unfortunately it was late - really late and performance wise it is similar to its litle X64 brothers. You see a future for it? When it is late again!!! What are you smoking?
"Mat Bryant...Simply labelling anyone that disagrees with your point of view a liar is not a very grown-up debating strategy"
LOL yeah. Neither is him constantly saying "slowaris" and "sunshiner" like some retarded pet parrot, nor claiming to track down other commenters' IP addresses, nor constantly droning on, and on, and on about HP, and Itanium, and all that rot...
His debating strategy isn't "very grown-up" because, the plain fact is, he isn't a grown-up.
@HANS on SAP
You are right Hans. SAP is crap seen from an IT perspective. But if things start to run MUCH slower when you upgrade, to a box that SUN have advised you to use. And the box you are upgrading from is 5 years old.
And it is not only SAP, to quote the comments that a former fellow student, which I have worked with when I was a consultant, had on Niagara based boxes. "It is crap that they sold me such a box and it is clearly not made for this kind of workloads" He tried to get a RAC cluster to perform on a handfull of T5120. He is one of the most Hardcore Solaris and Sparc fans I know, and also a very skilled DBA.
Now the truth behind these stories is that not all workloads are well suited for the Niagara based boxes.
And migrating a workload that have been written/tuned/setup for running with few threads, on a few fast processors is not always that easy. Not without doing a rewrite, reconfiguration and new tuning.
Not to mention locking problems, memory use for all those threads etc.
Niagara is *great++* for what it was designed for, but IMHO SUN is trying to sell it in where it doesn't fit.
@Jesper re: HANS on SAP.
We did workload comparisons, comparing 5240's to AMD quad CPU boxes. 5240's were made to perform better than quad AMD boxes (of a similar price level & quad core) but initially they didn't perform better.
Then the app vendor realised they were restricting the number of threads the box could run so opened it up from around 80 to over 350.
So yes, software runs well depending on workload but secondly, the software has to be capable decent levels of multithreading or yes, it could bottleneck. However, I don't believe SAP is incapable of being tuned to run in a highly multi-threaded mode and equally I don't believe SAP has any serial thread bottlenecking issue after all this time.
It sounds like your consultant may have been incapable of getting SAP or Oracle or both to use more threads, which sounds odd.
Just my tuppence.
PS: M(tw)att Bryant, your longwinded rants discredit your already shaky standing. Follow your own advice sometimes, stop trolling & ranting about how good HP is & people might listen to you, It looks like you want people to listen to you?
No one is interested in your marketing FUD which you deliver by the container load.
PPS: I hear Itanic is dying, or at least is being strangled quite hard : http://www.theregister.co.uk/2009/02/05/intel_delays_tukwila/
PPPS: Will you send us loads of photos of you crying & eating humble pie when Itanic & hence Superdome eventually die? :-)
RE: @Jesper re: HANS on SAP.
"....It sounds like your consultant may have been incapable of getting SAP or Oracle or both to use more threads, which sounds odd....." Or just that the server didn't suit the task in hand.
"...PS: M(tw)att Bryant, your longwinded rants discredit your already shaky standing...." Hmmm, seems my shaky standing is just fine seeing as all you Sunshiners seem to be unable to refute my points.
"...No one is interested in your marketing FUD which you deliver by the container load...." So real world examples and facts are now FUD? Seems like there must be an awful lot of anti-Sun FUD going on what with Sun tanking so hard.
"....PPS: I hear Itanic is dying, or at least is being strangled quite hard : http://www.theregister.co.uk/2009/02/05/intel_delays_tukwila/...." Try reading the article. Better still, try reading the one at http://www.theregister.co.uk/2008/11/29/whats_going_to_happen_to_sun/ as it maps out just what Sun's problems are. I suggest you worry about those. But then from your response I see you already are very worried.
"....PPPS: Will you send us loads of photos of you crying & eating humble pie when Itanic & hence Superdome eventually die?" I wouldn't bet on it if I were you. And it seems the market thinks the same - they're much happier betting on hp and Intel than Sun. Maybe they saw the market figures on how hp Integrity is taking share from Sun in the lucrative high-end, or maybe they just like companies that turn profits rather than bleed red ink. Either way, the evidence is stacked against you. Enjoy!
RE: @Jesper re: HANS on SAP
Well as for the SAP, I simply think that the problem was that the upgrade was to be a simple hw swap, so that there weren't put manhours aside for serious reconfiguration, and retuning.
And what angered the architect was that he would have liked to have known this up front, cause then he would have chosen a MXXXX box. I personally think that he should have known, that things would perform differently. Getting the user generated part of the workload retuned should be quite simple, and is most likely enough. But you never know if there is some abab code somewhere that doesn't scale with more threads.
As for the DBA that I went to university with, then it was a locking problem, that couldn't be resolved on the niagara box.And they will now prob' have moved the Database to either a MXXXX or I think they also have some p595's.