Archive for the ‘Sun SPARC’ Category
According to IDC’s latest server market share report, released earlier this week, Oracle is languishing in fourth place with 6.6 points of market share (IBM has 30.5 points of market share). When you consider that Oracle/Sun has lost server market share in each of the past seven quarters, you have to imagine that Oracle are desperate to stop the rot. Well, today our friends at Redwood Shores attempted to stem the tide by announcing a new TPC-C benchmark result for a cluster of SPARC systems. However, the benchmark result is far from impressive. Sure, the benchmark system has a huge throughput. However, it is woefully inefficient.
Today’s Oracle benchmark result uses 27 64-core Sun SPARC T3-4 servers to process more than 30M tpmC*. In contrast, IBM’s most recent clustered TPC-C result uses 3 64-core IBM Power 780 servers to process more than 10M tpmC**. There are many ways to look at this. You could claim that Oracle uses nine times the number of CPU cores to achieve only three time the performance. Alternatively, you could claim that each CPU core of the IBM system is able to achieve three times the performance of a CPU core in the Oracle system. Either way, in my opinion, it points to a very inefficient benchmark run. Such inefficiencies are surely a concern for customers who are paying for Oracle Database based upon the number of CPU cores in their systems.
At first glance, the cost efficiency of this new benchmark system from Oracle may appear to be impressive—their system costs 1.01 USD per tpmC. However, if you scratch below the surface, you will find that number is quite deceptive. Oracle do not use the perpetual licenses that you would expect, and Oracle do not use the kinds of support contracts that you would expect. If they did use the licenses and support contracts that are most commonly used, then the system costs would skyrocket, and the relative cost inefficiencies of this system would be plain for all to see. For prior coverage of Oracle’s price/performance tactics, see Sun and Oracle TPC Price/Performance Tactics Revealed.
Also, you should be aware that Oracle have once again resorted to sacrificing data integrity for performance in its benchmark systems. They have turned off page integrity checking—I imagine because, according to the Oracle documentation, it incurs a performance degradation of between 1% and 10%. So, even though it is highly unlikely that you would run a production system without page integrity checking, Oracle has chosen to do just that in the interests of squeezing extra performance out of its system.
Given all this context of misleading cost information and questionable system settings, it was timely to read the following article yesterday… Larry Ellison Hearsay: “We Can’t Be Successful if We Don’t Lie to Customers”
Results on Transaction Processing Performance Council Web site at http://www.tpc.org. Results as of 12/02/10.
* Oracle SPARC SuperCluster with T3-4 Servers (27 x 64 core) (108 chips, 1728 cores, 13824 threads); 30,249,688 tpmC; $1.01/tpmC; available 6/1/11.
** IBM Power 780 cluster (3 x 64 core) (24 chips, 192 cores, 768 threads); 10,366,254 tpmC; $1.38/tpmC; available 10/13/10
IBM today announced a new two-tier SAP Sales and Distribution (SD) benchmark result that crushes the Oracle/Sun results. The new benchmark result uses a 256-core Power 795 system with DB2 to support 126,063 SAP SD benchmark users. This is more than three times the number of users than Oracle’s largest system (a 256-core Sun SPARC Enterprise M9000 running Oracle Database). The IBM system was also able to handle more than three times the number of SAP SD users than Oracle’s result from September of this year that runs four clustered 32-core Sun Fire X4470 servers.
This new result is also the first to break the 500,000 SAPS level on a single system with more than 688,000 SAPS. The SAP Application Performance Standard (SAPS) is a measure of system throughput for business deliverables like customer sales orders or invoices. This result follows hot on the heels of last month’s result from IBM, which used a 128-core Power 795 system running DB2 to support 70,032 users. If you look at the effect of doubling the number of CPU cores from one benchmark to the next, you can see that the number of users supported is almost doubled. This near linear scaling is very reassuring for organizations whose SAP environments are growing significantly.
This benchmark result indicates that IBM can support your largest SAP systems, and that IBM can support the growth of those systems in a very efficient manner. But don’t forget that IBM Power Systems have tremendous virtualization capabilities. In fact, the latest optional PowerVM virtualization software allows customers to run more than 1,000 virtual servers on a single physical system. This means that you could also use a system like one in this benchmark to consolidate your various existing SAP environments onto one physical system, vastly simplifying your infrastructure and reducing costs in data center floor space, energy, management resources, and database and web application software.
Last year, I blogged about some issues regarding Truth in Advertising – Advanced Data Compression in Oracle 11g. Well, our friends over at Oracle are at it again. This time thay are making some questionable advertising claims. Here is the ad in question. From looking at the ad, you would think that an Oracle/Sun system gives you seven times the performance, while consuming one sixth the amount of power. Let me explain why this is MISLEADING.
There are two claims here, both under the banner of being “independently verified.” The independent verification refers to the fact that it is drawing from TPC benchmark data.
The first claim pertains to performance. When someone mentions performance to me in relation to the TPC-C benchmark, I immediately think of the primary metrics, and in tpmC in particular. After all, this is a primary metric for a reason. tpmC represents the performance of the systems for all workloads (for the record, Oracle did outperform IBM by about 20% for tpmC). But, Oracle obviously aren’t looking at tpmC when devising this claim. Instead, they are focusing only on one subset of the performance numbers in the benchmark. In other words, if the TPC-C benchmark were like a triathalon, then Oracle did really well in one of the events. It is downright misleading for them to claim that a 7x lead in one event is indicative of their performance in the overall race.
By the way, you should also be aware that Oracle are comparing an older IBM system to their latest and greatest, which is questionable in its own right. With the rate of change in the industry, IBM’s 2008 result is not indicative of its performance levels today. In addition, the Oracle configuration actually uses 115 TB of solid-state disk (for a database size of 6TB). The IBM result does not use any solid-state disk, instead working with mechanical disks. Solid state disk manufacturers claim that their products are hundreds of times faster than mechanical disk. However, for Oracle, that translated into only a 20% lead over IBM.
But, believe it or not, this is not as misleading as the second claim pertaining to energy consumption. First of all, the TPC results being touted here were posted before the TPC-Energy metric was introduced and reported. This energy data is not coming from the TPC results. Putting this claim under the “independently verified” banner is simply misleading.
Let’s dig a little deeper and do some math with the server specs. Note that Oracle needed a cluster of 12 SPARC Enterprise T5440 servers for their benchmark result, whereas IBM needed only one IBM Power 595 server.
If you go to the Sun SPARC T5440 Power Calculator, you can see that a single server consumes between 1551 watts (idle) and 2002 watts (100% active). There are 12 of these servers in Oracle’s benchmark, which results between 18.612 KW and 24.024 KW of power consumption.
If you look at the same information for the IBM POWER 595, you will see that during typical usage a P595 consumes 18.5kW. At 100% utilization, it consumes 27.7kW.
That’s right, the Oracle configuration in an idle state consumes more power than the IBM configuration performing a typical workload. Oracle, please explain how you arrived at the 6x number in the ad…
In case you are interested in the charts they were referring to, I will include a couple of them here.
The first chart compares two systems that achieve comparable performance for the 2-tier SAP benchmarks (well, actually the IBM system provides more then 15% better performance). In this chart, the focus is on the number of CPU cores needed to achieve these results. You can see that the IBM system requires 1/4 the number of CPU cores. This is important because software is typically licensed based upon the number of CPU cores, and therefore the efficiency of the system has a big determination on the price you end up paying.
The second chart compares how many SAP end users are supported per CPU core for IBM and Oracle systems. The per-core efficiency of the system is an important consideration for initial purchase, system maintenance, system upgrades, and system growth.
The first chart is based upon the following 2-tier SAP EHP 4 for SAP ERP 6.0 (Unicode) benchmark results, which are valid as of 4/7/2010:
- DB2 9.7 on IBM Power System 780, 8p / 64–c / 256–t, POWER7, 3.8 GHz, 1024 GB memory, 37,000 SD users, dialog resp.: 0.98s, line items/hour: 4,043,670, Dialog steps/hour: 12,131,000, SAPS: 202,180, DB time (dialog/ update):0.013s / 0.031s, CPU utilization: 99%, OS: AIX 6.1, cert# 2010013. For more details, see http://www.sap.com/benchmark.
- Oracle 10g on SUN M9000, 64p / 256-c / 512–t, 1156 GB memory, 32,000 SD users, SPARC64 VII, 2.88 GHz, Solaris 10, cert# 2009046. For more details, see http://www.sap.com/benchmark.
The second chart is based upon the following 2-tier SAP EHP 4 for SAP ERP 6.0 (Unicode) benchmark results, which are valid as of 4/7/2010:
- IBM SAP 2-Tier SD result of 15,600 SD (Sales & Distribution) users (Average dialog response time: 0.98 second), running DB2 9.7 on AIX 6.1 and SAP enhancement package 4 for SAP ERP 6.0 on the IBM Power System 750 with 4 POWER7 3.55 GHz processor chips (32 cores, 128 threads) and 256 GB main memory, certification Number: 2010004. For more details, see http://www.sap.com/benchmark.
- Sun Microsystems SAP 2-Tier SD result of 4,720 SD (Sales & Distribution) users (Average dialog response time: 0.97 second), running Oracle 10g on Solaris 10 and SAP enhancement package 4 for SAP ERP 6.0 (Unicode) on the SPARC Enterprise T5440 with 4 UltraSPARC T2 Plus 1.6 GHz processor chips (32 cores, 256 threads) and 256 GB main memory, certification Number: 2009026. For more details, see http://www.sap.com/benchmark.