Quicksearch
Disclaimer
The individual owning this blog works for Oracle in Germany. The opinions expressed here are his own, are not necessarily reviewed in advance by anyone but the individual author, and neither Oracle nor any other party necessarily agrees with them.
IBM's reaction to Sun/Oracle TPC-C announcement
Wednesday, October 14. 2009
Comments
Display comments as
(Linear | Threaded)
also compare response times of the top three results for power/sparc/intanic - sun is way in front here
#1
on
2009-10-14 16:57
I'm writing an article about some key takeways from this benchmark. The response time is part of it.
Hmmm... if Mrs. Stahl of IBM claims that the cores are 4.7 times more powerful, perhaps Oracle needs to raise the licensing fee per core on the IBM so it's 4.7x more expensive than the Sun?
I mean, even IBM is saying it's 4.7x more powerful... therefore to equalize the cost of licensing for the same work done.
I mean, even IBM is saying it's 4.7x more powerful... therefore to equalize the cost of licensing for the same work done.
#3
on
2009-10-14 17:31
Well, the results are 'In review', not yet accepted.
Salient points (imho) in the Executive Summary :
a. Why does the champion of Linux (IBM) use as 'other soft' Microsoft Visual C++ and COM?
b. So twice as much for rebate for IBM (-20,273,753) as for Oracle ($2,006,628)? That seems to me that IBM is giving their hardware(+maintenance) away? (cfr. subtotal page2)
Questions :
1. POWER 7 comes next year : does that mean that all the DDR2 memory is for the dustbin? ($2,602,598)
2. Last they told me that IBM still counts $1 = 1€ : does Sun/Oracle
do so too? If not what would the prices in Europe/rest of the world then be?
Elephant in the room : do I see anywhere the cost of the computer room? Which if in Tokio would dwarf the rest of the costs? (advantage Sun, no contest)
Congratulations Sun & Oracle!
P.S. Which entity has six or seven million users for one box? Thats nearly 20 times all the employees of IBM, isn't it? Just THINKing for myself
Salient points (imho) in the Executive Summary :
a. Why does the champion of Linux (IBM) use as 'other soft' Microsoft Visual C++ and COM?
b. So twice as much for rebate for IBM (-20,273,753) as for Oracle ($2,006,628)? That seems to me that IBM is giving their hardware(+maintenance) away? (cfr. subtotal page2)
Questions :
1. POWER 7 comes next year : does that mean that all the DDR2 memory is for the dustbin? ($2,602,598)
2. Last they told me that IBM still counts $1 = 1€ : does Sun/Oracle
do so too? If not what would the prices in Europe/rest of the world then be?
Elephant in the room : do I see anywhere the cost of the computer room? Which if in Tokio would dwarf the rest of the costs? (advantage Sun, no contest)
Congratulations Sun & Oracle!
P.S. Which entity has six or seven million users for one box? Thats nearly 20 times all the employees of IBM, isn't it? Just THINKing for myself
#4
on
2009-10-14 17:43
Errata :
So twice as much for rebate for IBM (-20,273,753) as for Oracle ($2,006,628)?
Ten times as much, decimally speaking of course. Doh!
P.S. six or seven : 5184000 of 6480000 users, I seem to have seriously rounded up the numbers here (people in meeting?), but the point is still valid. (Can't we knife this benchmark just for this reason alone : common sense?).
So twice as much for rebate for IBM (-20,273,753) as for Oracle ($2,006,628)?
Ten times as much, decimally speaking of course. Doh!
P.S. six or seven : 5184000 of 6480000 users, I seem to have seriously rounded up the numbers here (people in meeting?), but the point is still valid. (Can't we knife this benchmark just for this reason alone : common sense?).
#4.1
on
2009-10-14 19:04
I would like to kill it ... perhaps this get easier, now we lead it. But you don't believe how many people consider this as a good benchmark ...
Funny how fast IBM yell foul when they discovered someone else beat them using a clustered results. They themselves run a clustered results for TPC-H using 32 POWER6 p570 when they could have done it with a fully configured p595 (http://www.tpc.org/results/individual_results/IBM/IBM_570_10000GB_20071015_ES.pdf) , not to mentioned running it on AIX 5.3 instead of the so-called latest and greatest AIX 6.1.
#5
on
2009-10-14 17:50
One arm tied to the back? Isn't the Sun config fault tolerant, whereas the IBM config isn't? At least, that was mentioned by Larry during his keynote on Sunday...
IBM:
Winter is coming, do you feel the cold? The Sun is still shining.....
Winter is coming, do you feel the cold? The Sun is still shining.....
They will probably just add "non clustered" to the marketing material with a small font now, and use that from now on, just like she is doing in her post:
"...with IBM DB2 9.5 is the best overall non-clustered TPC-C system..."
"...with IBM DB2 9.5 is the best overall non-clustered TPC-C system..."
Keep in mind:
- IBM used 11,000 disks in their benchmmark for the server and storage
( 8 x 146GB SCSI disks and 10,992 73.4GB FC disks)
calculate:
- 5W per disks = 55,000 W for all disks
- then you have to add some power for the controllers assume 100W per controller - that's 68 * 100W = 6.8 KW for the disk controllers)
- next you have to power your Power server (wil be 20,000 W)
- adding this: 55 + 6.8 + 20 = 81.8KW
Now we know that the IT has also to be cooled - they didn't mentioned any water cooling in the FDR, so it will be conventional cooling technology - the rule of thumb is "1W cooling per 1W IT" - so you need another 81.8KW for cooling. In total that's 163.6KW for the IT and cooling. Assuming a price of 5 cent per KW/hour (very cheap) you have to pay 8€ (rounding down) for each hour you're using this equpiment. That's 192€ per day that will be 70,080€ per year and that will leave you with an energy bill of 210,000€ in three years.
- IBM used 11,000 disks in their benchmmark for the server and storage
( 8 x 146GB SCSI disks and 10,992 73.4GB FC disks)
calculate:
- 5W per disks = 55,000 W for all disks
- then you have to add some power for the controllers assume 100W per controller - that's 68 * 100W = 6.8 KW for the disk controllers)
- next you have to power your Power server (wil be 20,000 W)
- adding this: 55 + 6.8 + 20 = 81.8KW
Now we know that the IT has also to be cooled - they didn't mentioned any water cooling in the FDR, so it will be conventional cooling technology - the rule of thumb is "1W cooling per 1W IT" - so you need another 81.8KW for cooling. In total that's 163.6KW for the IT and cooling. Assuming a price of 5 cent per KW/hour (very cheap) you have to pay 8€ (rounding down) for each hour you're using this equpiment. That's 192€ per day that will be 70,080€ per year and that will leave you with an energy bill of 210,000€ in three years.
#9
on
2009-10-15 08:54
Joerg,
All this time, you whine about how TPC-C is a "terrible" benchmark. Now, Sunacle replaces HDD with Flash, which results in a whopping 26% increase over an IBM result with HDD (big surprise), and all of a sudden TPC-C is important.
Who are you kidding?
All this time, you whine about how TPC-C is a "terrible" benchmark. Now, Sunacle replaces HDD with Flash, which results in a whopping 26% increase over an IBM result with HDD (big surprise), and all of a sudden TPC-C is important.
Who are you kidding?
Sorry, but i wrote in every article about this TPC-C article, that i still think that TPC-C that tpc-c is horrible. Just read the blog entries. The day the TPC is retiring TPC-C is a day worth of celebration with wine and singing all night long.
From my perspective this result is important out of a single reason: Many customers still think that that TPC-C is a reasonable benchmark. In the years before this, we had to tell them "It's a nonsense benchmark." and the customers thought "They say this because they are bad in it". And when they saw, that we didn't reported a result they thought "Oh, they are really bad in it".
Now we can say "It's a nonsense benchmark." and the customer thinks "Ooops, they lead the benchmark, and still say it's a nonsense benchmark". And then we can talk about reasonable methods to compare the performance of architectures. That's the real importance of this TPC-C benchmark result.
Besides of it, it gives some interesting interesting insight about the load stability of some features in Solaris (like the FC target in COMSTAR)
If marketing has some opportunities to fire salvos in direction of IBM ... i'm fine with it, but in my perspective it's not the important outcome.
From my perspective this result is important out of a single reason: Many customers still think that that TPC-C is a reasonable benchmark. In the years before this, we had to tell them "It's a nonsense benchmark." and the customers thought "They say this because they are bad in it". And when they saw, that we didn't reported a result they thought "Oh, they are really bad in it".
Now we can say "It's a nonsense benchmark." and the customer thinks "Ooops, they lead the benchmark, and still say it's a nonsense benchmark". And then we can talk about reasonable methods to compare the performance of architectures. That's the real importance of this TPC-C benchmark result.
Besides of it, it gives some interesting interesting insight about the load stability of some features in Solaris (like the FC target in COMSTAR)
If marketing has some opportunities to fire salvos in direction of IBM ... i'm fine with it, but in my perspective it's not the important outcome.
I mind, what the opinion of Mr. Bluetothebones may be.
#11
on
2009-10-17 12:15
Well my point is, did Oracle/Sun did the apple to apple comparison.?
This new benchmark from Oracle/Sun is the same TPC-C benchmark that Oracle promised to reveal on October 14, which led to a reprimand by the Transaction Processing Council (TPC) for having made an unsubstantiated superior performance claim. (link to press release) In order to demonstrate "the world’s fastest database performance with a TPC-C benchmark" Oracle's used an Oracle RAC configuration (database for clustered servers) and a 12-Node Sun SPARC Enterprise T5440 cluster. It achieved 7,717,510 tpmC, with $2.34/per tpmC. Oracle compared this result to IBM’s single system 64-core IBM Power 595 with IBM DB2 9.5, based on a TPC benchmark submitted on June 10, 2008.
Here's a deeper look at Oracle's claim and the comparison to IBM's system:
• IBM system is a single server while the Sun system is a configurations with a 12 node cluster.
• The Sun cluster had a total of 384 processor cores and 3,072 threads compared to 64 cores and 128 threads in the IBM system.
• With 512GB of memory per node the Sun cluster had a total of 6TB of memory, compared to 4TB in the IBM Power 595 system.
• While the Sun tpmC per core is 20,097; the IBM tpmC per core is 95,080. The IBM result had 4.7 times higher performance.
• The Sun system will not available until 12/14/09. The IBM system has been available since December 10, 2008.
• The flash technology used in the Sun result was integral to the performance and price/performance. Flash technology wasn't even being used in benchmarks when the TPC-C published the IBM system result on June 10, 2008. IBM has leadership results using SSDs in benchmarks .
• For clustered systems, see the outstanding systems performance of IBM DB2 pureScale on Power Systems.
This new benchmark from Oracle/Sun is the same TPC-C benchmark that Oracle promised to reveal on October 14, which led to a reprimand by the Transaction Processing Council (TPC) for having made an unsubstantiated superior performance claim. (link to press release) In order to demonstrate "the world’s fastest database performance with a TPC-C benchmark" Oracle's used an Oracle RAC configuration (database for clustered servers) and a 12-Node Sun SPARC Enterprise T5440 cluster. It achieved 7,717,510 tpmC, with $2.34/per tpmC. Oracle compared this result to IBM’s single system 64-core IBM Power 595 with IBM DB2 9.5, based on a TPC benchmark submitted on June 10, 2008.
Here's a deeper look at Oracle's claim and the comparison to IBM's system:
• IBM system is a single server while the Sun system is a configurations with a 12 node cluster.
• The Sun cluster had a total of 384 processor cores and 3,072 threads compared to 64 cores and 128 threads in the IBM system.
• With 512GB of memory per node the Sun cluster had a total of 6TB of memory, compared to 4TB in the IBM Power 595 system.
• While the Sun tpmC per core is 20,097; the IBM tpmC per core is 95,080. The IBM result had 4.7 times higher performance.
• The Sun system will not available until 12/14/09. The IBM system has been available since December 10, 2008.
• The flash technology used in the Sun result was integral to the performance and price/performance. Flash technology wasn't even being used in benchmarks when the TPC-C published the IBM system result on June 10, 2008. IBM has leadership results using SSDs in benchmarks .
• For clustered systems, see the outstanding systems performance of IBM DB2 pureScale on Power Systems.
#12
on
2009-11-13 14:27
Did you read the article i wrote, or do you just copy over the IBM website here ? ( http://www-03.ibm.com/systems/migratetoibm/getthefacts/powersystem.html).
Of course it's an apple to apple comparision, because it's exactly the same workload. When you have a workload that is represented by TPC-C then the Sun system is clearly faster and cheaper than the IBM system. It does it differently, but thats irrelevant. I'Ve explained that in my article ....
Of course you can argue about TPC-C in it's entirety, but that another question.
To your copied text i stated already in the article you copied: "IBM tries now to tell the public, that this benchmark shows that Power is still superior, because they have more tpmC per Core. 4.7 times more tpmC per core - to be exact. I could calculate another derivate out of this - Price per Core in the configuration used for the benchmark: $18,051,719 / 384 cores = $47,009/core at Sun and $17,111,788 / 64 cores = $267,371/core for IBM. In the TCO calculation the single core at IBM is 5.6 times more expensive. This is the context you have to know, when Mrs. Stahl writes that a single core delivers more performance per core.
But both is totally irrelevant for this benchmark. What counts performancewise is the number of TPC-C transactions per minute. Period. When you want to proof the impact of a better single-core performance, use a different benchmark. The business of a customer is doing transactions in his architecture on his complete conglomerate of servers, storage and networking, not doing transactions on a single core. And the customer buy a certain number of cores to do the job, not a single core. For the customer the way to yield a certain level of performance is irrelevant. Cluster or single system? Irrelevant! Fewer fast cores or more slower cores? Irrelevant! Of course the customer must be sure, that he has chosen the right benchmark ... that TPC-C is representative for his or her workload, but that's the real problem of TPC-C."
I appreciate any discussion but please stop to copy IBM marketing in the comment field ...
Of course it's an apple to apple comparision, because it's exactly the same workload. When you have a workload that is represented by TPC-C then the Sun system is clearly faster and cheaper than the IBM system. It does it differently, but thats irrelevant. I'Ve explained that in my article ....
Of course you can argue about TPC-C in it's entirety, but that another question.
To your copied text i stated already in the article you copied: "IBM tries now to tell the public, that this benchmark shows that Power is still superior, because they have more tpmC per Core. 4.7 times more tpmC per core - to be exact. I could calculate another derivate out of this - Price per Core in the configuration used for the benchmark: $18,051,719 / 384 cores = $47,009/core at Sun and $17,111,788 / 64 cores = $267,371/core for IBM. In the TCO calculation the single core at IBM is 5.6 times more expensive. This is the context you have to know, when Mrs. Stahl writes that a single core delivers more performance per core.
But both is totally irrelevant for this benchmark. What counts performancewise is the number of TPC-C transactions per minute. Period. When you want to proof the impact of a better single-core performance, use a different benchmark. The business of a customer is doing transactions in his architecture on his complete conglomerate of servers, storage and networking, not doing transactions on a single core. And the customer buy a certain number of cores to do the job, not a single core. For the customer the way to yield a certain level of performance is irrelevant. Cluster or single system? Irrelevant! Fewer fast cores or more slower cores? Irrelevant! Of course the customer must be sure, that he has chosen the right benchmark ... that TPC-C is representative for his or her workload, but that's the real problem of TPC-C."
I appreciate any discussion but please stop to copy IBM marketing in the comment field ...
Yes I copied the from the same side cause you cannot measure 3000CC car with 3500CC car, for measuring you have to be on same level that’s the important point, I think you did’nt read the copied article that’s why I said “ITS NOT APPLE TO APPLE COMPAIRISON” and secondly your comparison and measurements are from Oracle side or if you want me to copy the side 
…
…
#13
on
2009-11-20 14:10
Yes I copied the from the same side cause you cannot measure 3000CC car with 5000CC car, for measuring you have to be on same level that’s the important point, I think you did’nt read the copied article that’s why I said “ITS NOT APPLE TO APPLE COMPAIRISON” and secondly your comparison and measurements are from Oracle side or if you want me to copy the side …
#14
on
2009-11-21 20:52
At first: I saw your comment this weekend, but as you were so eager to repost it, i've decided your article could need some maturing in the queue of articles that await moderation ;-> .... i'm not looking into that queue every day ... thus it can take a while until i'm approving comments on articles older than 30 days. This time limit helps me to reduce the level of spam ... as spammers often choose older articles.
At second: You didn't understand my point. I can do this comparision because it's an apple to apple comparision. It's an application benchmark. This benchmark tests architectures on their speed for solving the problem defined in the TPC-C benchmark rules. Of course i've read the article because it was the reason why i wrote the article you've commented at first. I would like to ask you, to think about this point from my article: " But both is totally irrelevant for this benchmark. What counts performancewise is the number of TPC-C transactions per minute. Period. When you want to proof the impact of a better single-core performance, use a different benchmark. The business of a customer is doing transactions in his architecture on his complete conglomerate of servers, storage and networking, not doing transactions on a single core. And the customer buy a certain number of cores to do the job, not a single core. For the customer the way to yield a certain level of performance is irrelevant. Cluster or single system? Irrelevant! Fewer fast cores or more slower cores? Irrelevant! Of course the customer must be sure, that he has chosen the right benchmark ... that TPC-C is representative for his or her workload, but that's the real problem of TPC-C."
TPC-C measures performance of an architectures on a certain workload, it doesn't test the CPUs. Of course IBM could just use multiple p595 but there is a problem with it and this is the only good thing about TPC-C, it includes a performance/price comparision.
At the end this discussion is useless, as TPC-C has left the realm of realistic workloads long ago. From my field perspective this benchmark just helps me to kill of TPC-C in the minds of customers. This is easier when you in lead the benchmark. Marketing may thing otherwise, but that the basic essence of my view to it.
At second: You didn't understand my point. I can do this comparision because it's an apple to apple comparision. It's an application benchmark. This benchmark tests architectures on their speed for solving the problem defined in the TPC-C benchmark rules. Of course i've read the article because it was the reason why i wrote the article you've commented at first. I would like to ask you, to think about this point from my article: " But both is totally irrelevant for this benchmark. What counts performancewise is the number of TPC-C transactions per minute. Period. When you want to proof the impact of a better single-core performance, use a different benchmark. The business of a customer is doing transactions in his architecture on his complete conglomerate of servers, storage and networking, not doing transactions on a single core. And the customer buy a certain number of cores to do the job, not a single core. For the customer the way to yield a certain level of performance is irrelevant. Cluster or single system? Irrelevant! Fewer fast cores or more slower cores? Irrelevant! Of course the customer must be sure, that he has chosen the right benchmark ... that TPC-C is representative for his or her workload, but that's the real problem of TPC-C."
TPC-C measures performance of an architectures on a certain workload, it doesn't test the CPUs. Of course IBM could just use multiple p595 but there is a problem with it and this is the only good thing about TPC-C, it includes a performance/price comparision.
At the end this discussion is useless, as TPC-C has left the realm of realistic workloads long ago. From my field perspective this benchmark just helps me to kill of TPC-C in the minds of customers. This is easier when you in lead the benchmark. Marketing may thing otherwise, but that the basic essence of my view to it.
Who cares how fast one core is? The entire CPU is the important thing. If my old Volvo car has a faster piston in the motor, does that make my old Volvo faster than a Porsche? No. You can not study one core, you must stude the entire CPU. No one buys one core, you buy an entire CPU. Therefore, compare cpu vs cpu. Not core vs core. Or, I might as well as compare ALU vs ALU, or register speed vs register speed, etc.
#15
on
2009-11-25 11:47
+1
The LKSF book
The book with the consolidated Less known Solaris Tutorials is available for download here
Web 2.0
Contact
Networking
xing.com
LinkedIn
Facebook
My photos
- mail: joerg@c0t0d0s0.org
- icq: 144000169
- Me, myself and I
- openBC
- My del.icio.us
- c0t0d0s0@twitter
- c0t0d0s0@last.fm
- wiki@c0t0d0s0
- Amazon Wishlist
Networking
xing.com
My photos
Comments
about Checking VXFS QIO usage
Wed, 23.04.2014 07:58
Joerg
Yes, ZFS is a diffe
rent animal. The part about di
rect IO and the POSIX locks ar
e true for UFS.
We ( [...]
about Checking VXFS QIO usage
Tue, 22.04.2014 08:56
Oh, i have some comments. Howe
ver as you set your homepage i
n the comment to Symantec.com
you may have more knowle [...]
about Checking VXFS QIO usage
Tue, 22.04.2014 00:25
Sorry, first a loooong explana
tion of QIO and then ODM and h
ow it will look and how to ena
ble it..
Long ago, wh [...]
about Checking VXFS QIO usage
Sun, 20.04.2014 11:27
This is exactly the problem. I
t seems even the way, that it'
s by default QIO and many peop
le think this is all the [...]
Buttons
This work is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 3.0 Germany License
Tracked: Oct 16, 10:08
Tracked: Jan 21, 08:28