Oracle, which will presumably soon own Sun Microsystems, has not said much about its plans for the Sparc server business at Sun since announcing its takeover in April - but apparently the top brass at the company are going to assuage everyone's fears that Oracle will ditch Sun iron, and will do so on October 14 at the OpenWorld …
TPC-C is so old and nothing like a real world app... its like saying look how quicly I can run Win 3.1 on this latest intel chip!!!! whoopy doo!
try TPC-E and give Microsoft a run for their money since they are the only ones submitting benchmarks in that category
TPC-C has merit - Funny Sun Agrees Again
There is no doubt in my mind the M9000 wouldn't stand a chance it barely beat the 595 on the light weight SAP SD 2-tier benchmark and core for core it wasn't even close. The thing that impresses me about TPC-C to this day, even thought the benchmark is basic in nature, is the sheer amount of disk attached to the servers to drive the I/O requirements for these big numbers. Just look at the disk behind the 595 - lots of it. Unlike many benchmarks, this one requires reads *and* writes and there is lots of changed data making its way back to the gaggles of disk spindles.
Anyone see JAVA's 4Q results? They need all the good news they can get over at Oracle. Sun's new sugar daddy must feel like dumping its new girlfriend's hardware division...
Whatever tpmC Niagara can do with RAC on TPC-C, I'd bet an x86 cluster or a POWER6 cluster could meet or exceed at a much lower price point.
Maybe Oracle will also announce they are going to price per socket to help?
Its still ironic that after all the TPC-C bashing Sun has done over the last decade they'd publish this result with Oracle.
With brutal memory bandwidth per core I don't think we'll see Niagara do TPC-H any time soon...
But it's a useless benchmark
Or so Sun have been saying since, oh, 2001 or so. They've been relentless in publicly saying this. I'll note that they aren't the only ones who say this but they've not been shy.
You may want to scale this up: http://www.oracle.com/ad/images/SUN_SparcvsIBM_2288sec.pdf
the math dictates.......
assuming this pic is the system and it is an existing rack mounted SPARC system, it should be a t5440. Which is a 4 RU server, max 10 servers in a 42RU system rack
Pic does not quite look like T5440, but close. perhaps they got rid of the internal disk slots or covered them up and went for SAN only.....
racks are max 42U high, there is 10 systems visible, and that would amount to 320 cores.
the oracle dbase price would be something like 320 X 6 - 10K dollars for oracle per year, assuming you know how to hustle your oracle rep ;-)
so, for the math impaired, you're talking about approx 2 - 3.2 million dollars per year.
the 64 core ibm box would check out for something like 450K - 700K a piece.
5 of those would give you approx the same # of cores, and a 25 - 30+ mln tpc benchmark......
question is: who needs it ( apart from the NSA echelon guys, of course....)
this article needs more oomph
sun servers are (in my 20 years of server room experience) the most solid and reliable, though generally slower and costlier.
as an admin, i really don't care how much crap my company can buy and stuff into the machine room. if they can save money, it's just going into someone's pocket, or being blown on something else.
since i have to admin retarded amounts of gear in the server room already, i will always prefer stability, support and manageability, over speed and price, any day.
which is why i prefer sun.
i hope larry keeps on cranking out good, reliable, slow, expensive sun gear. i'll still buy it, and see you all at the pub.
Oh now the fun begins!
First to the AC who yawned. TPC-E is *only* a microsoft benchmark because, if you believe the rumors, it allows for a more 'apples' to 'apples' benchmark for *hardware vendors*.
Yeah, I heard this from an IBM Information Management exec, so either he was telling the truth or just blowing smoke as an excuse why IBM SWG doesn't have a benchmark out there....
As to the article's estimates on the TPC-C running on Sun...
I think the estimates are well overblown. Not to pour water and douse Sun's er Oracle's marketing message, but that Oracle would have to do a major rewrite to take advantage on the additional threads per core.
It would be interesting to see the TPC-C numbers because IBM hasn't released an IDS (Informix) benchmark which is what they had earlier stated that it was their OLTP database of choice. (At least that's what they've said publicly at the IIUG user group meeting in Lenexa. Or was it IOD or both?)
Looks like we may be back in to the benchmark wars of the 90's.
So let the FUN BEGIN!
Mine's the FUD proof ballistic nylon flak jacket and helmet.
Expect Sun to drop the license cost of T2+ to .5
10 systems * 4 chip * 8 cores = 320 * .75 licenses = 240 Oracle licenses
Oracle EE + RAC + Partitioning + compression = 10.4M + $5M + $2.5M + $2.5M = $20.5M
Assuming you get 60% off (and if you don't get 60% off your getting ripped off) = $8.2M
And this is just for one year of maintenance.
Of course Oracle will show 1 year term licenses to pretend $47K / license is only $9.5K / License
Know anyone that buys term licenses? I hear they are only used for TPC.
Expect license trickery, solid state drives and trying to use the T2 which if you look at every Sun powerpoint was never made for databases. There is a good reason why Sun only propose's M-class for the database servers.
I would not be surprised if Oracle has the price for hardware at 150% off to sell all that software.
"I'll pay you to take our T5440 to run Oracle" Leisure suit Larry
Obviously the ad is in clear violation of the TPC rules, but I bet the fine they get from TPC will be less that 1/100th the cost of the front page ad.
The TPC should require them to do a retraction on the front page of the wall street journal
"We are sorry, we flaunted the TPC rules, we will try to pass off a web server as a database server in October and will pretend that people actually buy software on 1 year term licenses. While we may be committed to Sun SPARC hardware that does not mean we are not trying to sell it to HP ASAP so all of our maintenance stream does not dry up from the Oracle install base"
Get ready for Nehalem EX and Itanium Tukwila to have .75 license factors. 50% more cost for questionable performance per core. Its all about Larry and his boat and house built without nails.
....cheers from the UK
So 320 cores are faster than 64?
So SUN^H^H^HORACLE says 320 cores at 1,6GHz are faster than 64 at 5GHz?
Now who would have expected that?
And speaking of expectations:
I expect it will be only seconds before IBM pulls the magic "TCO" card, with 320 cores to license and ten boxes to put into maintenance, you´re in for several years of butt-pain. Even compared to the expensive (but good) IBM GS.
@exit....quit....bye.... yadda yadda
You very well know very well that 320 cores equals 40 Niagara CPUs. That is 40 octo core Niagara CPUs. Against 32 dual core IBM Power6+. It will be interesting to see which CPU is fastest. The 1.6GHz Niagara which "suffers from a small cache" or a 5GHz Power6+? If the Niagaras which "doesnt do for Database loads" according to FUDers here, achieve twice as high score than the Power6+ CPUs, then what do you say?
Anyway, I dont like the TPC-C benchmark. It is very artficial and no one uses such machines in real life. IBM's TPC-C machine costed 17 million USD and used 2TB RAM, and lots of Intel Xeons (no Power6 cpus). That is quite sick configuration and price.
SUN has said earlier that they dont like this bench and that is why SUN dont publish numbers on this. Also, the FUDers and Liars here, said that SUN doesnt publish numbers because SUN can not match IBMs numbers. I said that "if SUN wants, SUN can publish numbers on TPC-C, winning easily. But SUN is above the TPC-C benches, which prove nothing". But if Oracle (who are in command now) wants to show TPC-C digits, they can. Anyway, this bench only proves that the SUN machines are fast att doing TPC-C. I dont like this bench. It is too artificial and doesnt prove anything. (Well it proves that SUN can show TPC-C numbers if they want, proving that I am right. As usual).
@exit....quit....bye.... yadda yadda
"Anyway, I dont like the TPC-C benchmark. It is very artficial and no one uses such machines in real life. IBM's TPC-C machine costed 17 million USD and used 2TB RAM, and lots of Intel Xeons (no Power6 cpus). That is quite sick configuration and price."
What are you talking about Kebabbert? No Power6 cpus? You don't even have to open the .pdf disclosure to see some of the configuration information. It is on the webpage for the test, and clearly says Power6:
Result ID: 108061001
Result Status: Accepted
TPC-C Rev: 5.9
Report Date: 06/10/08
Total System Cost 17,111,788 USD
TPC-C Throughput 6,085,166
Price/Performance 2.81 USD
Availability Date 12/10/08
Database Manager IBM DB2 9.5
Operating System IBM AIX 5L V5.3
Transaction Monitor Microsoft COM+
CPU Type: IBM POWER6 - 5.0 GHz
Total # of Processors: 32
Total # of Cores: 64
Total # of Threads: 128
# of Clients: 128
CPU Type: Intel Xeon Dual-Core- 2.0 GHz
Total # of Processors: 128
Total # of Cores: 256
Total # of Threads: 1256
Regardless of how much you dislike TPC-C, haven't you ever looked at a result before?
You should not compare the two
There are many reasons you should not compare the two results
1) Single SMP vs. 10 system cluster
2) 64 cores vs. 320 cores
3) 128 threads vs. 2,560 threads
4) DB2 perpetual licenses vs. Oracle 1 or 2 year term licenses
5) Disk drives vs. new solid state drives
6) A processor that can do data base workloads and batch with a processor made for web apps requiring massive thread counts
7) Power an architecture with a future vs. SPARC which is on its death bed
8) A system with full hardware virtualizaton vs. 10 systems with thread partitioning
9) If you have read this far you understand the difference in architectures
10) If you have read all top 10 I bet you miss those wild and crazy top ten lists Scooter used to do....those days along with Sun are over....its an Intel and Power world now...the choices just got a lot simpler.
Cheers from the UK.
Kebabbert, I don't know how many years of IT experience you have.
But, as I remind in the dot com boom when Sun bought the remains of Cray design, they loved to do the TPC game, cause they had a bleeding edge technology with lots of processing power.
Well, then that cycle ended, HP had its glorious PA-RISC days well ahead of SPARC.
Suddenly, Sun started to hate TPC metrics... go figure...
At that time, Fujitsu was the only vendor to publish Solaris metrics.
Then, Fujitsu made an OEM agreement with zombie-Sun.
And, at that time, Itanium-2 and POWER5 were well ahead of Fujitsu.
Guess what? Fujitsu started to hate TPC... go figure...
TPC is a real 2 horses race for quite sometime. SPARC is a joke on database performance.
LDOM and Container are also horrible.
Solaris still solid and good, but with these hardwares and features, no one uses that anymore. Combine these with all European probe going on, who's gonna buy Sun boxes? At the end of probe, we will see a mere 1-digit percentage share for Sun.
"What are you talking about Kebabbert? No Power6 cpus? You don't even have to open the .pdf disclosure to see some of the configuration information. It is on the webpage for the test, and clearly says Power6:"
I was wrong on this, and you are correct. The fastest machine uses Power6. Yes, I did a mistake earlier when I instead looked at the "Client configurations" which uses intel Xeons and missed the "Server configuration". I skimmed that text too fast, as I dont like TPC-C and I dont consider a machine for 17 million USD realistic. Hence, TPC has no interest for real DB admins.
But you are correct, and I was wrong. I will never again say that the fastest IBM machine used Intel Xeons. Thanks for pointing that out!
(Compare my reaction to morons which will continue to claim otherwise, despite being proven he is wrong. Still he persists in claiming false things, and never changes his mind. If you point out an error I made, I immediately change my opinion and never state that error again. Correct facts are important. Never lie.)
IBMers are not going to like this. Oracle commits to SPARC and Solaris, says the website:
Oracle will spend more resources than ever. IBMers: rejoice! :o)
@ uh oh - kebabbert
never trust a sales -gag :-)
paris, cuz she figured that out a while before you did ;-)
- Twitter: La la la, we have not heard of any NUDE JLaw, Upton SELFIES
- China: You, Microsoft. Office-Windows 'compatibility'. You have 20 days to explain
- Is that a 64-bit ARM Warrior in your pocket? No, it's MIPS64
- Apple to devs: NO slurping users' HEALTH for sale to Dark Powers
- Apple 'fesses up: Rejected from the App Store, dev? THIS is why