Hopefully someone at Sun Microsystems can still explain why the Rock processor is still a good idea. Today, Sun confirmed what we told you long ago - it's delaying Rock-based servers until the second half of 2009. The company made this admission at its analysts conference in San Francisco. We didn't hear a single question from …
I thought Rock-based servers went out with the longbow.
Fowler has lost credibility
No Niagara server will match any RISC/EPIC server on database performance.
(except maybe the EOL UltraSPANCIV+)
Fowler is ignoring the fact that the Niagara cores do not provide single thread performance which is critical to database workloads. It reminds me of when Jonathan said the NiagaraII chip was 89.6GHz.
So how much does Oracle cost on a four socket NiagaraII box?
4*8*.75 = 24 Licenses required
24 * 40K = $960,000 + $192K/year maintenance
= the most expensive 4 socket server on the planet
"No Niagara server will match any RISC/EPIC server on database perform"
SPARC (and therefore Niagara) *is* a RISC server. So how does that work?
"So how much does Oracle cost on a four socket NiagaraII box?"
You can only have a one-socket Niagara-2 box, so your comment is flawed again. However, you can have a four-way RAC cluster using Niagara-2, which using Standard Ediition is per-socket, not per-core pricing. Furthermore, RAC is free under Standard Edition, so you can have a 32-core RAC cluster, with 128 threads for the cost of 4 Standard Edition Licenses.
= BY FAR! - The cheapest database server on the planet.
Research and facts are wonderful things - you should try them!
"4*8*.75 = 24 Licenses required
24 * 40K = $960,000 + $192K/year maintenance"
oh, and BTW, while we're on the subject of facts, the Oracle per-core pricing for Niagara chips is a 0.25 multiplier, not a 0.75.
So, in fact, an 8-core, 64-thread Niagara-II server will cost you two enterprise edition licenses.
So, still, the cheapest database server on the planet.
oh, and the UltraSPARC-IV+ is not EOL'd.
Did you get anything right?
Traditional RISC processor I should say.
According to Sun's server chief John Fowler, a four-socket Victoria Falls server will match....
4 sockets * 8 cores each * .75 licenses/core = 24
If you think pricing is only .25 check Sun's own benchmark disclosures.
T1 was .25
T1 1.4GHZ was .5
T2 is .75
Oracle Oracle Database 10g Release 2 Enterprise Edition, Per Processor 6**
Oracle Partitioning, Per Processor, Unlimited users for 3 years 6**
Oracle Oracle Database Server Support Package for 3 years 1
(Incident Server Support) 24x7x4
Oracle Oracle Application Server 10g Release 10.1.3.3, 6**
Java Edition, Per Processor
Oracle Oracle Application Server Support Package for 3 years 1
(Incident Server Support) 24x7x4
(** 6 = 0.75 * 8). Explanation: For the purposes of counting the number of
processors which require licensing, a multicore chip with "n" cores shall be
determined by multiplying "n" cores by a factor of .75).
People don't buy standard edition any more than they buy term licenses...well maybe 5% of customers
And Fowler was the one that said..."Plenty of Sun users to skip Fujitsu March"
Looks like Fujitsu has Sun by the short hairs now...and Solaris customers are still buying SPARC servers with a 150MHz interconnect....it pays to look inside and outside the box
An Adjudged Nudge in the Right Direction...... For Meteoric Meritorious Progress
"Why bother giving yourself an edge by arguably jumping a generation ahead of rivals with a superstar new chip? That would be too easy."
You think it would be too easy, Ashlee? :-)
MeThinks that Virtual Realisation only manifests ITself when the Quantum Leap is Taken and the Doors before You are no Longer there. ..... BeHold, Virgin PlaySpace and ITs Creative Spaces/Future Times.
And pray tell, what possible Good Reason would deny you a Generation Jump which can Lead a Follow/Follow a Lead which Leads into ITs Creative Spaces. Heady Stuff, I know, and thus all the More Reason to Boldly Go and take the One Small Step for a Man, One Quantum Leap for ManKind, Jump 42 KickStart the Engine.
And that's amfM Pinging the Sunshine State and Sunshine States ...4SunShineStates2 ..... and Unfurling a Blank Canvas upon which to Paint Life as Nature Intends and Nature's Inns Tend. Shared Visions of AI Beta Future.
"If Sun's figures hold up, then you have to wonder why it would even bother with Rock. Why not kill the chip and use the extra R&D money to keep the Niagara line on track?"
Ashlee, Rock can spend the extra R&D money, Niagara and Victoria Falls will/can so Dominate as to Generate their Own Crazy Diamond Flows. They already have their HyperRadioProActive Drivers and One All Ready, Born Right in the Heart of Bulawayo. Great Games Grand Wizardry, there 42 Control with Astute ProAction and SMART Investment/Inquiry .....for that appears to be the Default Control Game. :-)
Reading between those lines, Ashlee, and Rock 2 is, Delivered just in the Nick of Time by the Mights in Source in the Open. Transparent Network InterNetworking JAS. The NINJA Servers who like to say Yes.
Stepping up to that Base Plate Presents Virtual Control of the Future. I Kid U Not.
Who would like 42 Hazard a Lead..... make a Bold Statement ..... and Go Virtually NEUKlearer?
Sun's Up ..... Wanna Dance/Squawk/Make a Move, Sun Networks?
AI BetaTesting Fitness for Future Purpose.
Most expensive Oracle server
Oracle EE pricing for the T2 processors is 6 licenses / chip
Oracle was .25 for the 1.2GHz T1 processor and .5 on the T1 1.4GHz.
The T2 core is now priced like the USIV+ cores.
Oracle EE 6*40k + Oracle Partitioning 6*10K + Oracle App Server 6*$5K = $330K + 20% maintenance = $396K for the first year
The article was about Sun, not Oracle. If your pointy hair boss wants Oracle it might be worth of a comment in some other place but not here. How Oracle prices it products is not a Sun issue. For applications they work with IBM and DP2 and now they own MySQL so let's see on data base area.
Anyway - Sun is much like Xerox, many great ideas and even better implementation than Xerox but a little lacking how to commercial the idea. I think it will change. And Solaris is great, it is very easy, standards based, documented 30 yeras ago and still valid. For us, Linux users, they are doing OK. So, if they keep that way I think Sun will do well - just where is the Java (a joke, almost as .NET IMHO) word in Sun announcements? Do systems really need to accommodate developers and not users?
The answer is single-threaded performance. There are commercially important applications which do not parallelise well, and for these there really is no substitute for something which will run a single-thread as fast as possible. Rock aims to provide that.
RE: facts ....
I'll try and keep this really simple for you. Niagara is a castrated RISC design, with all the guts ripped out to make it as slim as possible so as many as possible can be squeezed onto one die. Look at the transistor count for a core compared to even UltraSPANKEDIII, Sun had to rip out half the workings to get the shrink job done. The result is a simplified core that is great for simple small threads like webserving, but just cannot handle real database tasks, unless you completely rewrite the whole database to a massively parallel design (I bet the MySQL team are trembling in their boots at the thought!). To put it simply, any Niagara, even factoring in the Sunshine fantasy vapourware predictions, will be trounced in an Orcale shootout by a three-year-old Xeon, let alone Power6/7 or Tukzilla.
Niagara Any-Number will be the most expensive Oracle server ever for the simple reason that, after you have wasted time and resources on evaluating it, you will then have to go and buy another vendor's box. And as to how much more FSC can wring out of SPARC64 econimically is debateable. That is why Sun has to get Rock going, or it can kiss goodbye to the real enterprise space where the margins are richer and the services opportunities are plenty.
FSC systems no longer a stop gap
Well, the SPARC64 based systems themselves had come out late, and the concern was that users would delay purchase because Rock wasn't far behind. This latest reschedule clearly makes these systems a "destination" rather than a stopgap. And they certainly deserve to be a destination in their own right, making a natural progression from much of the USIV+ systems.
Matt is right about the single threaded performance of Niagara, but we already know that the duty cycle of the typical database installation includes *both* heavy OLTP (parallel) *and* batch processing (single threaded). Devising an effective way to meet these disparate requirements is hard whatever processor architecture - there is little more straight line speed to be had anywhere.
Whatever, with SPARC64 systems, Niagara in single sockets now and dual/quad sockets later in the year, and the x86 range, Sun has a comprehensive portfolio with or without Rock. The "this screwdriver is no good for banging in nails" commentary is unhelpful.
How Oracle prices it products...
Tuomo: "How Oracle prices it products is not a Sun issue"
You sir are mistaken. Larry's Oracle licensing practices are arguably one of the biggest logs on the Sun dethroning camp fire of the early part of this decade.. People don't by Sun - they buy solutions. And if a Sun / Oracle / App solution costs twice as much as a technically inferior but otherwise equally as capable Sun / IBM / App solution (because Larry's licensing scheme says so) - they wont buy Sun and they will buy IBM. As they did. In the zero sum game of Unix Server.
"Castrated RISC design"
This is the best comment I've ever read, for those of us that remember the first RISC systems. How did we ever manage without our "evaluate polynomial" instruction, I wonder?
This might come as a shock, so digest it slowly ...
Not. Everyone. Runs. Oracle.
Rumor has it that there are even sectors out in industry that don't care about traditional DB performance.
Matt, Matt, Matt
I was going to comment on what an idiot Matt Bryant is, but thought about it a bit more. Matt appears to be quite intelligent... The problem is that he knows juuuust enough to be dangerous. My guess, based upon his obvious distaste for Sun and all things Sun, is that he works for one of Sun's competitors (probably IBM). No facts, just FUD.
Look at the Niagara I/Oracle performance numbers on OLTP and you will see some very impressive numbers. Look at Niagara II/Oracle performance numbers and you will see even better OLTP numbers. Sun is always cheaper than IBM, especially when you take into account IGS (and you always have to take that into account). Yeah, Oracle licensing is an issue, but Oracle supports Solaris Containers so these boxes are a superior consolidation solution. Of course DSS isn't as good, but there are other solutions for that. Niagara wasn't designed for DSS...
Is this a correct piece of info, that the Niagara systems
are just $285 million (I believe that is just "bookings" also)
I would think that for a company like SUN, a successful product line would
have at least $1b in bookings.
I mean, they only have so many product lines.
here's something else to consider
apart from license cost of whichever product:
A fsc or 25k can have multiple boards combined which means that the domain can restart after a hardware failure. Downtime tops 10-20 minutes, usually far less.
On a fsc there is (like powerpc since power3 ) even processor failure protection up to a reasonable degree; a core failure does usually not bring the OS or application down if you have more than one.
In contrast with that, the rock-based planned systems are single chip or at least single backplane. Failure equals "buy new box and wait" or force you to have a cluster. (==double license cost)
I do not see banks, stock traders etc run high value per 1 transaction applications on that kind of kit. Not in the shops I know, anyway.
There is no multidomain like you have it on 25k planned or foreseen for Rock based servers. The single board design is a design choice by Sun
and imho it invalidates the rock for monolithic high value processing environments. And these are still commonplace.
What I am seeing happening, is that they turn to SUN (built by Fujitsu/Siemens) SPARC64 boxes because of the much better RAS they offer. Or they turn away altogether and shift to IBM p570 (1 - 4 systems in a single system, always recovers no matter what)
RE: Matt, Matt, Matt
Aw, a fan at last! I feel honoured, but I have to admit to not working for any vendor, but I do have to suffer regular doses of Sun FUD so I think it's only fair I get to spit back every now and again. In fact, seeing as you obviously have the attention span of a goldfish (and the goldfish intellect), you have obviously forgotten the comments where I made fun of IBM.... And Dell.... And EMC (well, not much of EMC as their salesgrunts are a bit "committed" and tyres are expensive).
I have looked at Niagara performance with Oracle, and to be frank, it was pants (don't whatever you do go read tweaker.net, you'll only get upset!). Niagara2 came out with an FPU per core per still sucked at Oracle, and was still too expensive compared to x86-64 to make sense for anything other than webserving. But Sun had already prepared us for that, they said "Wait for Rock, but have some nice SPARC64 in the meantime." The fact that Sun have had to go with SPARC64 and Xeon/Opteron is BECAUSE Niagara is too niche to make it as a real enterprise platform. If Niagara was the be-all-and-end-all chip they wouldn't have needed to perform the world's biggest volte face and get all lovey with Intel and Mickey$haft, would they?
Niagra and Rock are good at different things.
Niagra Boxes are really very different boxes from the things we are likely to see from Rock Boxes.
Niagra (1 and 2) will do lots of things very quickly but you wouldn't want to sit anything truly mission critical on top of a single box it's a 1u or 2u box with (maybe) one redundant psu and a software mirrored boot disk - you need to build redundancy in at a rack level using good design in software. Rock boxes are reported to be medium (2 socket) to large systems (8+ sockets) on different system boards and will likely have the ability to be split up into multiple domains configurations like the old enterprise 2500 and the current fujistu/sun m9000 boxes are. The sort of redundancy and ability to address huge amounts of ram you can get out this sort of configuration is going to be used for different tasks like big number crunching operations on huge high end database applications - there is plenty of room in the market for both sorts of box, and I'm sure some people will buy both for and use them for different purposes. It's about having a choice!
ROCK in mid 2006!
Amazing what you find on the web...check this out
You know the web never forgets....which would make Sun a habitually tardy
The road map has rock in 2006, USIV+ continuing, USIIIi+ not canceled yet and Intel Xeon Systems without the 3 year hiatus
Me thinks Folwers pants might be on fire.
RE: Matt, Matt, Matt
Face it Matty, there is no denying that you have an anti-Sun agenda written all over you. Show some facts:
"I have looked at Niagara performance with Oracle"
- Where is the data, may be you have looked at your own crooked shelf and found nothing to get excited about !!
How much ever you deny of not working for any Sun competitor, your posting history, from your apathetic Sun attacks, devoid of facts of course, doesn't support it. Most likely you are an HP reseller, and your day to day pay depends on the commission you get by selling HP kits.
I have a simple suggestion for you, Stop your anti-Sun FUD, before you loose total credibility here.
Jump ....the Water is Deep but only if you cannot Swim
"Rumor has it that there are even sectors out in industry that don't care about traditional DB performance"............. Ye Gods, the Bounders? Ye Gods, the Blighters? Ye Gods, the Jumpers?
And I would like to dedicate this post and launch IT on its way to Chief-in-Jumper-in-Chief, Sir Richard Branson, who may also be into "Controlling Systems ......... with Crack Code Communications 4Core2Core I/IO" ...... http://comments.theregister.co.uk/2008/02/05/ibm_bluegene_web/ ....... as he considers what he could do with a 50 billion guaranteed loan to Invest for the Future.
Oh that one was so occupied, you may say, but it is a task very few would even attempt. Being forever the Romantic, the White Knight Flying in on his Own Space Craft to say Hello and No to Hello, is the Stuff of Magic Legend which always delivers Gold, Frankincense and Myrh. [Frank in Sense and Demure]
Re: $285 million?
$285 million per quarter, not per year.
RE: RE: Matt, Matt, Matt
Sorry, wrong answer, would you care to guess again? I don't work for HP or a reseller. I do have an anti-Sun agenda because I have suffered previously from Slowaris poisoning - it's really annoying watching a project you know works fine already on hp-ux being gradually hacked to junk because some Sun salesguy came in with his pants round his ankles and told the management "We can do it cheaper, faster, better!" only for the truth to be "It's not stable, we can't get it to integrate, it runs like a three-legged dog on Slowaris". I know it's hard for you Sunshine fanatics to realise but there are people out here that would rather use hp-ux, Linux or Windows (or even that nasty AIX stuff) rather than touch Slowaris on SPARC, Niagara or x86, and we're enjoying watching the zig-zag progress of Schwartz & Co as they desperately try and keep their hardware bizz alive.
And to all the Oracle doubters out there, whilst LAMP is the stack for the edge, when you get into the real enterprise space then it's Unix and Oracle supporting the big apps like SAP. Oracle performance is very important to many big corporations as most of their bizz critical apps will have an Oracle database (or a comparable product like Siebel or DB2) included as part of the stack, which is why Sun need Rock.
RE: RE: Matt, Matt, Matt
So you are one frustrated guy out there to take revenge with all vengeance. Some Sun sales guys wanted to sell you something and it didn't work - big deal !! Your porting project fails and you start putting the blame on somebody - talk about pointing fingers...
Here is an interesting story for you to read:
I guess these guys are gunning for IBM, just like you.
As far your Slowaris moniker, may be you are too much out of date. This term was applicable to pre-s10 x86 version of solaris, not any more.
Yes, Oracle is important for big iron enterprise stuffs - but when it comes to Oracle RAC on these tiny servers, LAMP can be a great replacement. Of course Oracle wants to push RAC more than their regular 10g/11g because they can charge for both per cpu and per node and get more money - MySQL has the potential to be a great money saver. It's just a question of time - Linux was only on the EDGE few years ago, today it is encroaching into mission critical areas. So Larry comes up with his own Larry Linux to make more money. The monetary incentives are just too hard to ignore for open-source databases to go the same way.
Sun doesn't need ROCK today, any more than HP needs Tukwilla. The 4-core upgrades to Fujitsu servers will keep the clock ticking for years. Who knows whether promise of ROCK will ever materialize, I can't care less. I just like to see Sun filling up their x86 portfolio faster with lot more Intel based x86 servers and lot more supported free software.
RE: RE: RE: Matt, Matt, Matt
So, only people that say nice things about Sun are allowed to post here? Sorry, I didn't realise this was the Sun Fanclub website! Thanks so much for clearing that up, but I think I'm about to upset you again.
"Sun doesn't need ROCK today...."??? Reality check - Wintel/Lintel is a very competitive arena and Sun is very much Johnny-come-lately to a part of the market where margins are skinny and services restricted by the fact many customers can do the job themselves. However, the enterprise space (where companies pay more to have the solid performance, support and scalability of big iron UNIX) is margin-rich, offers great services opportunities, and is worth more for less physical servers units than the Lintel/Wintel bizz. Even with their growing capability, neither Linux nor Windows on x86 can truly replace those big iron units today (well, maybe on Itanium they can), which means Sun has a problem. Sun can't survive on just the Lintel/Wintel bizz it has, it would have to massively scale back if it accepts being the number four x86 vendor, and at the same time would need to widen it's x86 portfolio as even Dell has a broader offering. Do you think any Sun CEO wants to go to Wall Street and say "Hey, guys, we used to be the number one UNIX vendor but now we just want to be a 2nd tier Dell clone?" I think not. Sun desperately needs its customers to believe in Rock as otherwise those customers will go to another vendor and probably another commercial UNIX. FSC and the APL give Sun a breathing space, but expect IBM and HP (and Red Hat, SuSE and Mickey$haft) to try and pinch the Sun customers. At the moment, Sun's flip-flops and slippages are making it easy for them - IBM has a solid and believable roadmap, so does HP, but Sun's has looked a shaky for years....
And yes, MySQL is a great money saver, on the edge - in the datacenter core, customers pay for support for bizz critical apps, which puts MySQL right in there against Oracle, DB2, etc, etc. Look at Linux on iSeries or Integrity, it all comes with proper support at a price. If a customer is going for freeware, what's to stop them using PostgreSQL, or even the freebie SQL built into Windows Server 2003? Sun have a lot of work to do before MySQL becomes the golden goose you seem to think it is, and given Sun's record with software acquisitions I'd not be counting on it.
Where growth is
> So, only people that say nice things about Sun are allowed to post here? Sorry, I
> didn't realise this was the Sun Fanclub website! Thanks so much for clearing
> that up, but I think I'm about to upset you again.
Of course you are entitled to your opinion, how much ever one sided they may be,
and the same is applicable to me. I have argued with your post because you are
on the other side of the spectrum - the anti-Sun/pro-HP (who seemingly pretend
to be pro-IBM/DELL as well, just to hide the pro-HP stance) zealot - feel free to prove otherwise.
Yes, I repeat that Sun doesn't need ROCK today, because Fujitsu/APL covers that
space very well for the next two years. Delay in getting a chip out is not new in the non-x86 area - but because of the sticky nature of this business, customers make low purchasing decisions and then they stick for a very long time, and they don't migrate unless options run out on them - and I don't see delay in ROCK roll-out
as one reason why a Sun account would consider migrating. Sun has to deliver
on ROCK to prove on their commitments, but a chip delay with another very
viable option present isn't alarming to Sun to say the least. Sun has been selling
the US-IV+ based servers even now, even though they are utterly uncompetitive,
and they aren't selling them bad either.
You are right, Sun can't survive on x86 business, but that's the area which is
growing. Unix business isn't growing - it's slowly shrinking because some
applications are migrating to x86. HP/IBM/Sun are scrambling to get the piece
of the debris, but nobody is betting their business growth on it. x86 is where the
growth is coming from. Where do you think Oracle gets most of it's revenue these
days - from the Unix DB licenses or from the x86 licenses - feel free to point to
appropriate data instead of hand waving. Many of these x86 Oracle licenses can
instead use MySQL and would save lots of money even with 24x7 support. There
are lots of markets outside the Fortune-500 mission critical area which need
reasonably good databases at the best price - MySQL has a popularity lead in this space compared to PostgreSQL. And Windows bundled freebie SQL ? May be
it's good for a hobby project, these sexy new startups are all doing Linux -
windows isn't on their radar. A customer starts with free MySQL on Linux, but
as it's operation becomes bigger, his inhouse support is no longer enough -
tha't when is opts for paid support on the same software and can still save money
compared to keeping an inhouse support team.
And of course Sun's new x86 servers are not second tier DELL clone, not unless
you are working for HP or DELL marketing. Margins on x86 servers are slim on
x86 server hardware, but not on software/service/support. If Sun can execute well
on their software side, they can get a lot more margin compared to DELL.
So yes, when Sun CEO goes to the wall street and begs in front of them, that's
because he has seen that the ground has already moved from underneath Sun - all those Unix customers has already migrated to much cheaper
x86/Linux/Windows boxes - and the CEO already knows that these customers will
never move back to another Unix box again, whether IBM or HP or Sun for these
applications already migrated. Guess what ? More and more customers keep
buying a lot more x86 servers compared to Unix servers for their data crunching
AND these web applications, and more startups are coming up completely based
on Linux/Opensource and keeps growing lot faster than these old horse business
houses. It's wise to try to grow by courting these geeks while trying to keep the
revenue stream from the enterprise backbone steady.
Tell me about how much HP and IBM are poaching Sun Unix customers with
their Unix gears (not migration to x86) because Sun doesn't have ROCK (I count
IBM is doing it well, but I very much doubt HP). I bet Sun would be happy
defending themselves with APL and put the extra energy behind their x86
business and software.
- NASA boffin: RIDDLE of odd BULGE FOUND on MOON is SOLVED
- SOULLESS machine-intelligence ROBOT cars to hit Blighty in 2015
- BuzzGasm! Thirteen Astonishing True Facts You Never Knew About SCREWS
- Worstall on Wednesday YES, iPhones ARE getting slower with each new release of iOS
- Tor attack nodes RIPPED MASKS off users for 6 MONTHS