Oracle and Sun taunt IBM with Sparcs If Oracle keeps saying that it wants to be in the server and storage hardware business long enough, maybe some people will start believing it. Ahead of its Oracle OpenWorld customer event, which is being held from October 11 through 15 in San Francisco, the company has been running advertisements that poke a stick at IBM, …
"For what Oracle is outlining in this advertisement to be true, Sun had to have already nuked its Sparc and Solaris development efforts starting last year and continuing through the period when Sun started shopping itself around to IBM, Oracle, and presumably also Hewlett-Packard and Fujitsu."
That doesn't really follow, or make sense to me, can you explain what message it was intended to convey ? ('Sun must have stopped R&D if Oracle is doubling the effort".... don't think so...)
Author writes, "Billings for Sparc Enterprise machines fell by 54 percent in the fourth quarter ended in June, Sparc T machines were up a tiny bit, and x64 server sales dropped by 9 percent."
This is a pretty clear indication where the market's confidence is - in Sun developed Open SPARC processor based servers.
I am not sure what can be done to boost the sagging market confidence of the Sun/Fujitsu Enterprise servers.
All of the other proprietary server manufacturers were seeing declines during this period - even the servers leveraging proprietary Intel & AMD CPU's.
I don't think I agree with "poking IBM with a stick". I would not necessarily want to see their products "go the way of the dinosaur" - competition from multiple vendors is healthy for the market and healthy for the consumer.
So Soreacle will release a TPC bench highly reliant on a flash array to get the figures? And exactly how long do you think it will take IBM to come back with their own mega-bench figure by simply putting a layer of SSDs into the same kit it used for the last test? I'd really laugh if IBM let Soreacle get half-way through their day and then released their own killer flash TPC figures at a higher number than Larry's.
SPARC systems will be low low prices until they are gone gone gone.
Most of the Sun sales force has been let go, relying on partners.
Just counting the Oracle account execs the sales force will "double"
The Tclass is not made for database workloads regardless of what oracle might be able to do with 10 boxes and using TB's of cheap memory sticks vs disks and spindles.
Oracle is all about profit and their goal is to turn the $500m losing sun into a $1.5B profit engine. That money will come from customers, not from new sun hardware which takes 5+ years to invest and see new products.
Rock is dead Larry. Is your claim before or after that compare since you have not admitted its death yet?
DB2 now has Oracle compatibility....me thinks that what you are really scared about.
What I got from this is read the ad....call IBM for their view
I've seen a Sun roadmap for the SPARC that's still under wraps, but basically kills off the Rock (which we knew) and moves some of that team to develop the T2 strain further.
In general, Sun's technical marketing appears to have gone to the mattresses. They won't talk to me without an NDA and I don't feel like signing one right now.
Most glaringly absent from Larry's announcement is any mention of x86 and StorageTek. Sun's x86 range, especially the blades, is a generation behind IBM and HP. But what will they do with all those hypervisors?
IBM just mopped the floor with Sun - see IDC 2Q UNIX revenue data. Larry had to say something! Let's look at the past few Oracle and Sun quarters to shed some light on this...
Oracle to spend more money than Sun on SPARC - Oracle has money, Sun is losing money all the time...tough one there...Oracle can afford to boost R&D, Sun can't. More money for Solaris...ditto. Oracle to tune the stack - I can't believe that's not going on already - heck, the Oracle on Power slide decks I see from IBM showed me that 40 or so people show up for work at Oracle tuning Oracle and AIX together. After all, who won all those database benchmarks including Oracle Apps? Oh yeah, IBM AIX on Power gear.
Oracle with more sales reps than Sun? Really? Goes back to that making money thing...who would have guessed...
If Niagara is the only branch in the family tree (good thing Sun bought Afara) not being hacked off...I guess they better stop calling it a network facing processor - databases were supposed to go on UItraSPARC processors right? Oh yeah...Sun cancelled V and VI and started putting their plastic covers on Fujitsu servers...
Sugar Daddy Larry just put a nice big diamond ring on the thin bony finger of a used up old tramp! As soon as you marry her Larry, take her out on your yacht and push her overboard - act sad and phone your life insurance company! Then, keep the kids you like from her family, but that hardware kid is ugly - put it up for adoption - maybe Fujitsu will feel sorry for it and give it a home!
I think Larry has seen in the last few quarters less and less profit from SUN and is worried about the next few SUN quarters where more customers will be going somewhere else until the EU regulators are done screwing things up.
To me, the add is like a hail marry pass that might work and satisfy some IT shops, but my IT shop has already put plans on migrating off of SUN equipment (upper management says we can't buy any new SUN hardware ... only new/used parts). I wish that the StorageTek gear was never purchased by SUN since I'm looking at replacing that gear with another vendor. For the SUN servers, they will be migrated to AIX or Linux on HP gear depending on the situation.
Matt is right.
Any vendor can make a clustered benchmark and get a 'big' result. The reason why the Vendors stopped doing these mega clustered benchmarks is that, we the customers, could see it for the bull it was.
And if Soracle opens this 'battle front' again they will just get their butts kicked.
I mean for example a distributed DB2 database is something that has been around for a long time, it is even a shared nothing database so it will scale really really well.
So for example lets say that IBM did the following:
1 x 42 inch rack.
4 x 9U Bladecenter each holding
14 x JS23 Blades each holding
4x 4.2GHz Power6+ cores.
Each blade will do about 350-400K TPMC, based upon it's rPerf rating. And lets just be seriously pesimistic and say that the scaling factor i 95% on a shared nothing database.
That would put one rack in the range of 14-19 Million TPMC.
Or Imagine a rack full of HP half hight blades with Xeons with for example DB2 ?
This is a can of wuuuppa** that Soracle shouldn't open.
// Jesper
This post has been deleted by its author
IBMers said that Larry would kill of Sparc, but let us see what the future holds. As I have said earlier, the Power6+ CPU can not compete with the new Niagara design on certain server workloads. It is too slow. The Power6+ relies on having a big and fast cache, but no matter how big, the cache will never be able to fit in a whole server workload with many clients. The only way forward, is to build a new design which is not dependent on a cache - i.e. Niagara SPARC. That is the reason Niagara can have a small cache and still crush CPUs that has much higher frequence and much larger cache. Large cache and high frequency is obsolete, Legacy.
Many cores are the future. Which IBM has always smacked SUN for. I wouldnt be surprised if IBM will turn to many cores instead of a few cores trying to reach 6-7 GHz. That is just a dead end. I hope IBM will realize it soon. But maybe not.
.
And please, stop using the number of cores when comparing CPUs. Instead, say that the TPC-C IBM record used 32 CPUs. If Niagara can break it with fewer CPUs then Niagara is faster.
Who says that the nr of cores is the important metric? It is subjective. You could as well use the frequency instead. "IBM has the TPC-C record with 4.7GHz CPUs". That doesnt say anything. Neither does "IBM has the TPC-C record with 64 cores". Maybe it is one CPU with 64 cores? That CPU would be awesome. It is more correct to say "IBM has the TPC-C record with 32 cpus" - when you want to show that a CPU is faster than another CPU. If you say "64 cores" then you can claim that IBM has faster cores - but that doesnt mean that the entire CPU is faster. Please stop that FUD.
"And please, stop using the number of cores when comparing CPUs. Instead, say that the TPC-C IBM record used 32 CPUs. If Niagara can break it with fewer CPUs then Niagara is faster.
Who says that the nr of cores is the important metric? It is subjective. You could as well use the frequency instead. "IBM has the TPC-C record with 4.7GHz CPUs". That doesnt say anything. Neither does "IBM has the TPC-C record with 64 cores". Maybe it is one CPU with 64 cores? That CPU would be awesome. It is more correct to say "IBM has the TPC-C record with 32 cpus" - when you want to show that a CPU is faster than another CPU. If you say "64 cores" then you can claim that IBM has faster cores - but that doesnt mean that the entire CPU is faster. Please stop that FUD."
Number of cores is still very important for software licensing, especially Oracle. Software costs are much more than hardware costs.
Agree with Jesper.
The last and only entry on the TPC-C clustered benchmark is more than 6 years old and if Oracle decides to reopen that contest they will be flattened by the competition.
RAC is totally over selled. While Oracle has made it scale to a few nodes in a clinical lab environment, RAC blows under real workloads. Going from one to two nodes will cost you twice the hardware and three times the Oracle license. And the result will be a decrease in performance unless you have hit jackpot with a workload that coincidentally matches RACs flawed design.
And in the real world a RAC cluster is a house of cards. In practice planned and unplanned downtime is much greater than with a one node setup.
http://mysite.verizon.net/vzet9sxl/YouDoNotNeedRAC.pdf
While the p595 in the TPC-C benchmark is an off-the-shelf highly flexible system that will perform excellently under a variety of real workloads, a 10 node RAC of 5440s will be something that will made for the TPC-C test only. Maybe you will find one of those systems in Frankensteins garage but not in a real data centre under real workloads.
Last time I checked Total cost of ownership was more important than technology definitions.
Oracle prices on cores not the chip. Two M5000's in an Oracle cluster with EE, RAC, partitioning and compression requires 48 licenses.
The software bill is $4.1M! Even with the standard Oracle 60% off the cost is $1.6M.
Q. Why talk about the M5000 only being "8 processors" when you need 48 software licenses?
A. If you are Sun and want to hide the software costs until after the poor performance hardware is shipped.
In 2004 McNealy begged Oracle to price per socket on stage at Oracle world. Oracle responded by buying BEA and moving them from socket to core based pricing. Sun is gone because of Larry's core based pricing and insane .75 multiplier for SPARC which only delivers .2 of performance per core.
Look at T2+ Oracle pricing....that is the worst in the world.
Cheers from the UK
"While the p595 in the TPC-C benchmark is an off-the-shelf highly flexible system that will perform excellently under a variety of real workloads,"
That's a completely outrageous statement! If you don't think IBM "game" the TPC-C benchmark themselves your living in dream-land, just take a look at the full disclosure report...
http://www.tpc.org/results/FDR/TPCC/IBM_595_20080610_fdr.pdf
Look at the section that covers how the DB2 database was set up - there's 628 pages of DDL/SQL stuff on how they partitioned the hell out of their data so transactions fit neatly into the cache lines on a P6 processor!
Compare that to the HP submission for 4M TPM:
http://www.tpc.org/results/FDR/TPCC/tpcc.hp.SD.fdr.082107.pdf
That runs to about 27 pages of DDL/SQL.... (and thats still too much!)
Now the last time you setup a database, how much DDL/SQL did you need to crunch to do it? My guess is that for something as straightforward as the schema that makes up the TPC-C benchmark, you might have a couple of pages worth... now you start to realise just how much IBM "fix" this benchmark. Don't believe me? Well why not go to the horses mouth and see what IBM's own techies say about the benchmark:
http://www.sigmod.org/sigmod/record/issues/0506/p71-column-winslet.pdf
The IBM 6M TPC-C system is as much a frankenstein creation as Sun/Oracle's new one will be - it all just goes to show that _all_ these benchmarks are vendor games (and I'd extend that to the SAP and SPEC benchmarks as well).
Folks need to wise up on this stuf and realise that:
- All (IBM Power, Sun SPARC, Intel Itanium, Intel x64) current CPUs provide "good enough" performance for 99.99% of all workloads - system performance isn't the issue it used to be - system *utilisation* (and TCO) is what folks should be focussed on - In case you hadn't notice we don't all drive dragsters!
- If you need some performance metrics, benchmark your own app - don't trust anyone elses numbers
Some fun stuff for thought.
1) ROCK is dead, all development cancelled, all release cancelled.
2) Initial RK systems were being shopped as >7KW and >14KW with extremely high thermal output. They would have been completely uncompetitive in an efficiency focused market.
3) Niagra and Niagra 2 are not as good as marketing leads you to believe. Or anywhere near it. Unless an application is written very specifically for the architecture and processor, they will underperform UltraSPARC-IV by as much as 70%.
4) Niagra/Niagra2 are NOT suitable or usable for databases, and Sun has known this since day one. You will be told to go pound sand if you complain of performance problems.
5) Enterprise M8000 and M9000's were significantly more expensive than comparable HP and IBM gear, based on pre-merger quotes. Core-for-core pricing on some was as much as double.
6) People are not migrating systems the way Sun claims or wants; E2900's go to M8000's, because M5000 lacks RAS. Even if performance is there, RAS is not, and it is a requirement in most environments.
7) Enterprise M8000 and M9000 operating costs exceed Superdome and P595 by tens of thousands per year. Cooling, power and facility requirements are drastically higher as compared to HP or IBM offerings. Additionally, M8000 and M9000 cannot be installed on non-raised flooring at all - nor can they be installed on standard raised flooring. (You can verify it with publicly available design documents, if you like.)
8) Core-for-core (that is, 1 core is 1 core is 1 core,) SPARC64 systems past the M4000 are physically larger, require more power than most other systems, and cost significantly more to operate per year. This gets much worse when you involve feature matching or performance matching.
9) Sun has and continues to have a severe customer confidence problem with Solaris 10, arguably justifiable. This is part of why you have Solaris 8 and 9 branded containers. Many large customers have steadfastly refused to go to 10 for numerous reasons. Whether or not 10 is better, that's not what the customers think.
10) Sun prior to shopping themselves around did make significant and drastic cuts in all departments. By simply relabeling Oracle sales staff as Oracle+Sun, they more than double sales.
11) Sun has been and continues to suffer from severe brain drain, mostly quietly. Large numbers of developers and experienced support staff were let go or resigned. I personally had a ticket that went through four people in a 3 month period, because the prior person "no longer worked at Sun." The Oracle sale predictably accelerated this.
12) Sun is much more dependent on OpenSolaris and external unfunded resources than publicly admitted. Pre-merger, Sun admitted to me that development resources were almost exclusively break-fix. New features and functionality were expected to be "provided by the community."
13) The customer dump, when customers under several million per year were thrown to the wolves - I mean VARs, was not executed. It was simply done with very little warning, and absolutely no coordination or organization. Customers were told to go find their own VAR. Two VARs I spoke with spoke bitterly of the lack of information and support from Sun. One suddenly found themselves handling an account with four E15Ks and no qualified in-house staff.
14) Sun support has become not even a joke, but rather, a horror to be avoided at all costs. I have not spoken with anyone to date, who is not utterly disgusted with Solaris support at this point. Most have reached the point where they will avoid it at all costs. All stated without qualifiers that they found better answers and support from OpenSolaris forums, every time.
15) I've been a Sun customer and user since 1993.
16) For many years, Sun did offer a vastly superior product with vastly superior service.
17) For many years, Sun was the leader in research and development.
18) For many years, Sun was justifiably regarded as a pillar of stability and reliability for both hardware and software. Especially software.
19) For many years, Solaris was justifiably regarded as the most advanced Unix available.
20) Those years have been over, and there is little hope that Oracle can reverse this, especially with the customer-hostile handling of things. Failing to respond to rumours and inquiries clearly, if at all. Complete lack of roadmaps or guidance on hardware. Price increases on software support across the board.
They have to counteract not only the severe problems they inherited that were Sun's own doing, but also a very strongly negative perception and reaction from the remaining customer base. Justified or not, that's the perception of customers and the image Oracle's presented.
1: Probably but until a day ago Matt B was predicting with almighty authority that all Hardware was for sale to HP. Are you as good at telling the future as he is?
2: Thats pure speculation considering no official specs have even been provided. Go look at Superdome : 24k max, 19k typical consumption. IBM P595 max is 27k as well. So assuming even an element of truth in those figures the Rock platform looks reasonable unless I missed something?
3: Utter bollox, we've been doing tons of migrations of Sparc iii, iv and iv+. No such issue though you can only take my word for it. Don' think my employer would let people in for a nosey around.
4: Every DB is the same? All perform badly? Lets wait and see what Oracle produce on the 14th and see if it really does have tons of SSD or cache bolted in. Even if it does have it that can be perfectly valid if a customer can replicate it in-house and make something fly in-house.
What do you reckon happened when people first started using caching Raid arrays? Did they say "Invalid, it's cheating using new technology, an unrealistic system spec......"
The 10/11g RAC technology isn't exactly new and I bet pricing per core changes pretty soon.
5: Having never quoted for Superdomes vs P595's vs M-series & factored in support, software and purchase cost I can't comment. I doubt you have either so smells like fud.
6: E2900's have to go to M-series? Yeah, right, seen a vendor promote that recently for their product but T5240's were easily as comparable. Core speed is different, yes but it copes with the IO a lot better which more than compensates for it. SSL offloading has to ease CPU as well which probably accounts for a fair few CPU cycles on other CPU's.
Anyway, I give up, can't be arsed to write through every other item though point 9 is laughable, 11 has an element of truth, 14 could be said of many places (IBM offshored Mainframe support, reckon that went well?, Spoke to an offshored guy for MSA arrays and I think he'd just learned how to spell MSA).
Hey ho, thats the way the world spins, spit FUD FUD & FUD again and a few will start to believe it.
It's a valid sales technique I suppose.
"While the p595 in the TPC-C benchmark is an off-the-shelf highly flexible system that will perform excellently under a variety of real workloads..."
WAHAHAHAHAHAAAAHHHHAAAAA!!!!!! LOL! LMAO!!!
That's hilarious...
10,992 15K 73GB Drives. That's like 600% more storage than required for the test. I'm sure everyone does that much more than required. 17Million for IBM's setup. I guarantee that the Oracle/Sun setup will be a fraction of that cost.
Also, I thought it was against TPC rules to compare Clustered results to UnClustered results...
Maybe not... but if Oracle is doing that, then they should be ashamed. Oracle quotes the unclustered result in their ad so I wouldn't think so.
"Number of cores is still very important for software licensing, especially Oracle. Software costs are much more than hardware costs."
Yes of course. But still, please stop talk about the nr of cores and instead talk about the nr of CPUs.
1. If IBM really wants to claim that Power6 is faster than Niagara, then IBM has to compare CPU vs CPU. IBM can not compare core vs core, that is plain wrong.
2. It is far more interesting to know how many CPUs are used, when talking about energy savings. 64 Power6 cores = 32 CPUs. Each 4.7GHz Power6 CPU uses maybe 500 Watt. This totals 16.000 Watt.
Compare this to 64 Niagara cores = 8 CPUs. Each 1.4GHz Niagara uses 105 Watt. This totals 840 watt.
Isnt it important to know that IBMs choice is 16.000 Watt vs SUNs 840 Watt? If you talk about cores, this is not clear. IBM wants to hide this power savings, and therefore tries to shift focus to talk about cores instead of CPUs. Also, IBM knows Power6 looses big time when comparing CPU vs CPU, therefore shifts focus to cores.
Therefore, please dont talk about cores. It is misleading. Talk about sockets. Oracle will reprice the licensing to not punish Niagaras. 8 core Niagaras will be much cheaper. And crush all competition on certain work loads.
Hp did not announced the roadmaps BEFORE the aquistion was finalised. I was working at hp when Carly did her UDI and introduced the Compaq piranas and as usual the product announcements could only be done after finalisation. Therefore that line of reasoning is just BS!
As to the FUD of having more people selling Sun after the Sun has gone down - before the announcement Sun had 32k resources and Oracle +80k employees? I will leave the sum making to you.
And looking at the Oracle roadmap - they are not interested in the big iron anyway - they want to play in the little league where they can get some leverage from their RAC Everything strategy but has been struggling to get the linux msg to stick.
Where I come from a core = processor = CPU they all just put more of them on shared silicon these days, but whatever.
Kebabbert - a 500W processor? You are out of your tree!
Where I come from we also run our databases on systems, so I care more about the total system power consumption - the CPU really doesn't matter that much in the scheme of things. The system still has to power the memory and the fans and the PCI slots and the SAN storage needs power - blah blah blah - you see where I'm going by now I hope.
Where I come from we also pay for our database by the core count - so we do care - very much.
When Oracle finally announces a TPC-C benchmark we can all look at if for performance, price, price/performance and performance per watt, or whatever angle we want to look at it.
Let's not lose sight of the fact we are talking about Oracle - and Oracle is priced today per core. While that will have to change for any SPARC processor to be competitive, that's the way we roll today.
Paris - because she might might also believe in a 500W processor...damn that was funny...
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
