BM Seer Unofficial thoughts from an anonymous Sun employee

SAP-SD Sun SPARC Enterprise T5440 WR 4-Chip

Monday Oct 13, 2008

The Sun SPARC Enterprise T5440 server with four 1.4 GHz UltraSPARC T2 Plus processors achieving 7520 SD benchmark users on the two-tier SAP Sales and Distribution (SD) standard SAP ERP 6.0 (2005) application benchmark.

The Sun SPARC Enterprise T5440 server with four 1.4 GHz UltraSPARC T2 Plus processors achieved over 37,000 SAPS.

The Sun SPARC Enterprise T5440 server with four 1.4 GHz UltraSPARC T2 Plus processors showed 80% improvement compared to Sun SPARC Enterprise T5240 with two 1.4 GHz UltraSPARC T2 PLUS processors.

SAP does not allow any SWaP or price-performance comparisons to be made.

The Sun SPARC Enterprise T5440 server with four 1.4 GHz UltraSPARC T2 Plus processors outperformed all 2-chip and 4-chip systems.

The 16-core 16RU IBM p570 POWER6 used four times the space of the 4RU Sun SPARC Enterprise T5440 system.

The Solaris Operating System, used in this benchmark, clearly demonstrates that the system can support over seven thousand concurrent users accessing the SAP application server.

This benchmark was run with Oracle 10g. Oracle 10g performed very well on the Sun SPARC Enterprise T5440 server.

SAP-SD 2-Tier Performance Table (in decreasing performance order, Benchmark Users, bigger is better)

System Users Chips,
Type
GHz		 Mem OS DB LI/Hr SAPS BM rev Date
IBM p 570 8000 8 DC POWER6 4.7 128 GB AIX 5L Ver 5.3 DB2 9 801,330 40,070 6.0 21-May-07
Sun SPARC Enterprise T5440 7520 4 US T2 Plus 1.4 128 GB Solaris 10 Oracle 10g 753,000 37,647 6.0 10-Oct-08
HP ProLiant DL580 G5 5155 4 6C Xeon 2.66 64 GB Win- dows Server 2003 EE SQL Server 2005 516,670 25,830 6.0 12-Sep-08
HP ProLiant BL680c G5 4432 4 6C Xeon 2.4 64 GB Win- dows Server 2003 EE SQL Server 2005 443,670 22,180 6.0 12-Sep-08
Sun SPARC Enterprise T5240 4170 2 US T2 Plus 1.4 128 GB Solaris 10 Oracle 10g 418,000 20,900 6.0 09-Apr-08
IBM p 570 4010 4 DC POWER6 4.7 64 GB AIX 5L Version 5.3 Oracle 10g 402,330 20,120 6.0 21-May-07
HP ProLiant DL585 G5 3801 4 QC Opteron 2.5 64 GB Win- dows Server 2003 EE SQL Server 2005 380,330 19,020 6.0 11-Jul-08
HP ProLiant BL685c G5 3524 4 QC Opteron 2.3 64 GB Win- dows Server 2003 EE SQL Server 2005 353,000 17,650 6.0 28-Mar-08
IBM p 550 3104 4 POWER6 4.2 64 GB Redhat Enter. Linux 5 DB2 9.5 312,670 15,630 6.0 29-Jan-08
Sun SPARC Enterprise T5120 2175 1 US T2 1.4 64 GB Solaris 10 Oracle 10g 219,000 10,950 6.0 09-Oct-07
HP ProLiant DL385 G5 2102 2 QC Opteron 2.3 32 GB Win- dows Server 2003 EE SQL Server 2005 210,330 10,520 6.0 28-Mar-08
HP Integrity BL860C 1165 2 DC Itan2 1.66 24 GB HP-UX 11i v3 Oracle 10g 117,000 5,850 6.0 22-Sep-08

Complete benchmark results may be found at the SAP benchmark website http://www.sap.com/benchmark.

Benchmark Description

The SAP Standard Application SD (Sales and Distribution) Benchmark is a two-tier ERP business test that is indicative of full business workloads of complete order processing and invoice processing, and demonstrates the ability to run both the application and database software on a single system. The SAP Standard Application SD Benchmark represents the critical tasks performed in real-world ERP business environments.

SAP is one of the premier world-wide ERP application providers, and maintains a suite of benchmark tests to demonstrate the performance of competitive systems on the various SAP products.

Disclosure Statement:

Two-tier SAP Standard Sales and Distribution (SD) standard SAP ERP 6.0 (2005) application benchmark: SPARC Enterprise Model T5440 (4-way, 4 processors, 32 cores, 256 threads) 4 x 1.4 GHz UltraSPARC T2 Plus, 128GB memory, 7520 SD Benchmark users, 1.98 sec avg response time, Cert#2008058, Oracle 10g, Solaris 10; SPARC Enterprise Model T5240 (2-way, 2 processors, 16 cores, 128 threads) 2 x 1.4 GHz UltraSPARC T2 Plus, 128GB memory, 4170 SD Benchmark users, 1.97 sec avg response time, Cert#2008021, Oracle 10g, Solaris 10; SPARC Enterprise Model T5120 (1-way, 1 proc, 8 cores, 64 threads) 1 x 1.4 GHz UltraSPARC T2, 64GB memory, 2175 SD Benchmark users, 1.91 sec avg response time, Cert#2007059, Oracle 10g, Solaris 10; IBM System p 570 (8-way, 8 processors, 16 cores, 32 threads) 8 x 4.7 GHz POWER6+, 128GB memory, 8000 SD Benchmark users, 1.98s avg resp time, Cert#2007039, DB2 9, AIX 5L Version 5.3; IBM System p 570 (4-way, 4 processors, 8 cores, 16 threads) 4 x 4.7 GHz POWER6+, 64GB memory, 4010 SD Benchmark users, 1.96s avg resp time, Cert#2007038, Oracle 10g, AIX 5L Version 5.3; IBM System p 550 (4-way, 4 processors, 8 cores, 16 threads) 4 x 4.2 GHz POWER6+, 64GB memory, 3104 SD Benchmark users, 1.91s avg resp time, Cert#2008002, DB2 9, Redhat Enterprise Linux 5; HP ProLiant DL580 G5 (4-way, 4 processors, 24 cores, 24 threads) 4 x 2.66 GHz Intel Xeon, 64GB memory, 5155 SAP SD Benchmark users, 1.97 sec avg response time, Cert#2008050, SQL Server 2005, Windows Server 2003 Enterprise Edition; HP ProLiant BL680c G5 (4-way, 4 processors, 24 cores, 24 threads) 4 x 2.4 GHz Intel Xeon, 64GB memory, 4432 SAP SD Benchmark users, 1.99 sec avg response time, Cert#2008049, SQL Server 2005, Windows Server 2003 Enterprise Edition; HP ProLiant DL585 G5 (4-way, 4 processors, 16 cores, 16 threads) 4 x 2.5 GHz Quad-Core AMD Opteron, 64GB memory, 3801 SAP SD Benchmark users, 1.99 sec avg response time, Cert#2008041, SQL Server 2005, Windows Server 2003 Enterprise Edition; HP ProLiant BL685c G5 (4-way, 4 processors, 16 cores, 16 threads) 4 x 2.3 GHz Quad-Core AMD Opteron, 64GB memory, 3524 SAP SD Benchmark users, 1.98 sec avg response time, Cert#2008016, SQL Server 2005, Windows Server 2003 Enterprise Edition; HP ProLiant DL385 G5 (2-way, 2 processors, 8 cores, 8 threads) 2 x 2.3 GHz Quad-Core AMD Opteron, 32GB memory, 2102 SAP SD Benchmark users, 1.99 sec avg response time, Cert#2008015, SQL Server 2005, Windows Server 2003 Enterprise Edition; HP Integrity BL860C (2-way, 2 processors, 4 cores, 8 threads) 2 x 1.66 GHz Dual-Core Intel Itanium, 24GB memory, 1165 SAP SD Benchmark users, 1.94 sec avg response time, Cert#2008052, Oracle 10g, HP-UX 11i V3; SAP, R/3, mySAP reg TM of SAP AG in Germany and other countries. More info www.sap.com/solutions/benchmark.

Results Summary

Certified Results
Performance: 7520 benchmark users
Server: Sun SPARC Enterprise T5440
Processors: 4 x 1.4 GHz UltraSPARC T2 Plus
Memory: 128 GB
Operating system: Solaris 10
Database S/W: Oracle 10g
SAP S/W: SAP ECC 6.0
SAP Certification: 2008058
Storage: 3 x Sun StorageTek(tm) 2510
3 x Sun StorageTek(tm) 2540
2 x Sun StorEdge 6120

[11] Comments
Like this post? del.icio.us | furl | slashdot | technorati | digg
Posted at 01:47PM Oct 13, 2008 in Benchmark Issues
Tags:
Comments:

Wow, that really is impressive and I take back all I said about T2 not being competitive with Power chips. I guess it's down to your very cool Zambezi crossbar with great bandwidth and low latency. Do you have any more recent IBM p570 and/or IBM p560 benchmark comparisons that March 2007? You've got to admit that was quite a while ago.

Posted by Kevin Hutchinson on October 13, 2008 at 02:55 PM PDT #

IBM has NOT updated the IBM p570 4.7GHz Power6 and submitted any new results. IBM has also not published any IBM p560 SAP results.

I fear we won't see anything new. When Sun publishes results, IBM often avoids comparisons.

Posted by BM Seer on October 13, 2008 at 03:03 PM PDT #

IBM did submit a new p570 result:

http://download.sap.com/download.epd?context=40E2D9D5E00EEF7CDD644A3C68D2556961C005FA302CAB8CD34EE06F2BED5C2B

32 cores - 14432 users

Posted by Thu Nguyen on October 13, 2008 at 05:54 PM PDT #

hmmm 32-core IBM p570 is 32 RU and how many $M? That is half of the largest IBM POWER6 system made. So you are telling me that it is faster than Sun's 4RU box. How much faster? How many more $? How many more watts (measured only please)?

Posted by BM Seer on October 13, 2008 at 05:59 PM PDT #

The 32 core p570 is 16RU, not 32RU. Maximum power for site planning is same as the 16 core p570.

Posted by Thu Nguyen on October 13, 2008 at 06:19 PM PDT #

OK, that's a funny game calling both p570. So Thu Nguyen you seem to refer to 4.2GHz p570/32 with QCM (still Dual-core chips) 256GB of 667MHz dimms (8RU).

IBM estimates the p570/32 at 5600Watts(max is higher according to IBM docs), so I'll assume 5000watts. Wow. likely not good power/performance on benchmarks that allow that, Again this is even with the reduced speed CPUs.

"IBM is only offering 4.2 GHz Power6 cores in this new Power 570 box, and that's due to heating, cooling, and power draw issues, according to Ross Mauri, general manager of IBM's Power Systems division." IT Jungle.

By the way it still seems to be a cabled together cluster like the other p570s, what is the off node latency compared to on-node latency. Note IBM does call them nodes - I guess even they think it is a node-cluster.

What is the price of this beast? Can you get one as configured above for $200k or is it double or triple that? Do the slow 400MHz dimms reduce the price much at 256GB? How much performance is lost with 40% slower memory?

Still kinda big at 2x the RUs of T5440.

Posted by BM Seer on October 14, 2008 at 09:51 AM PDT #

So a new IBM p570 with 16 processors (32 cores, 64 threads) does 2x more SAP users (14400 versus 7520) than a Sun T5440 with 4 processors. But the p570 is also 4x the size and draws about 3x the power. So the T5440 still stands up as the best SAP system around, especially looking at price/performance. Very impressive!

IBM Power 570, 16 processors / 32 cores / 64 threads,
POWER6, 4.2 GHz, 128 KB L1 cache, 4 MB L2 cache per core,
32 MB L3 cache per core, 256 GB main memory

Posted by Kevin Hutchinson on October 14, 2008 at 11:16 AM PDT #

BM Seer. IBM says, "In reality, if you just looked at even one comparison, in this case using SPECjbb2005, you would see that the IBM Power 570 is over 2 times the performance of Sun's new T5440."

Please see: http://www-128.ibm.com/developerworks/blogs/page/benchmarking/20081013

Please comment on this comparison

Posted by 15.243.169.73 on October 14, 2008 at 12:42 PM PDT #

Hilarious! The IBM audacity to confuse people with core count and avoid cost or wattage comparisons, wow. They'll say anything then try to make it cutesy.

How about the truth: So the IBM 32-core power6 system that IBM main performance blogger (Elisabeth Stahl) talks about is a 16RU 32-core systems that is half the core count of IBM largest Power6 server! What does that system cost Over $300,000USD when configured with memory and activating everything? how many actual measured watts during the benchmark? near 5000 watts?

Why are you asking this from an IP address associated with HP?

Posted by BM Seer on October 14, 2008 at 01:52 PM PDT #

... and yet the 570 sells like hot cakes and we all expect the new 32-way 570 to do the same. Remember the 570 is an enterprise server - the 5440 just another low end box from the bewildering arrays of hardware offerings from Sun.

Posted by Investa Ixnotsun on October 19, 2008 at 07:47 AM PDT #

Sun beats IBM on system-wide comparisons, in other words find comparable systems (based on server-size, server-price,...) then compare. Sun wins.

IBM can't even compare on chips which are something you can physically see (By the by, Sun's are more powerful), but they must compare on core count. They don't tell anyone that IBM's cores are ultra-expensive and very low density (it take 16 Rack Units to put 16 cores in a Power p570!).

I don't think p570's are selling like hotcakes I think they are *Canabalizing* iSeries. Let's run the numbers shall we...

If you look at the IBM numbers below are some scary facts, now notice IBM doesn't have clear reporting and IBM changes its "accounting" for pSeries and other servers to cloud the numbers. But if you take the time to
investigate...

Power Series (pSeries) +7%
iSeries down -82%

IBM has merged the old pSeries and iSeries into the new p-series. So, this way moving forward, the new Power-pSeries will show improvements over last year (due to the addition of iSeries business).

It is safe to say that IBM's p-series likely was flat to lower Vs last year. And, xSeries (x86) business was way down (-15%) including a drop in their Blade business.

Alex gives sweeping generalizations, I show the numbers. I also take to customers that are switching. In the end IBM can probably find customers that are switching the other way in this big world.

Let's contrast this with Sun and the CMT (before the T5440 and its great performance like you see in the posting that started this):

From Sun's press release: "Sun reported 83 percent year-over-year billings growth in its Solaris-based Chip Multi-Threading systems as customers continued to demand the nearly 10,000 applications available for Solaris 10, while enjoying integrated virtualization and exceptional power efficiency."

Posted by BM Seer on November 04, 2008 at 09:00 AM PST #

Post a Comment:
Comments are closed for this entry.