Oracle Development Unleashed (3rd Edition)

Page 1301

• Instruction Set Computation (CISC), and RISC may also utilize Branch Prediction computation on supported CPUs. A common example includes the wildly popular 275 MHz 64-bit DEC Alpha processor. DEC currently has a 64-bit Alpha processor running at 1000 MHz.

• PowerPC: The PowerPC chip is half as thin as the Pentium and therefore more efficient in power consumption, electrical resistance, and computational speed. PowerPC chips are also expected to be the next popular, true 64-bit processor introduced in late 1995. The PowerPC architecture is newer , more flexible, requires less overhead, and can process up to six simultaneous instructions per cycle versus the current two to three instructions of Intel 86 P-5 and P-6 high-end processors. The high-end Intel 86 requires roughly twice the number of transistors per chip to maintain the backward compatibility of previous 86 CPUs.
• Smart Controller Cards: Smart controller cards enable significant price/performance ratio increases among the smart card providers, and they are currently utilized only by the Oracle RDBMS. This is the result of Oracle's high-speed record latch and network management research. Smart cards offload computation from the main processor's CPU. A smart card is usually placed directly at the cable interface to the file server. In the AT&T Global Information Systems 3000 (formerly the NCR 3000) platform, smart cards attach to the host adapter, before the network wire and eliminate network protocol exchange between the network and the CPU. This is accomplished by means of a secondary processor available from the smart card. The resulting efficiency can be as large as a 50 percent performance throughput increase for a total investment of less than $5,000.00.

Making the Right Choice

With all these MP, SMP, and MPP platforms, how can you be certain to make the right choice for your business applications today and tomorrow?

Transaction Processing Performance Council (TPC) Benchmarks

The Transaction Processing Performance Council benchmarks effectively illustrate :

• Simulated banking application benchmarks on SMP via the TPC/A, B, and C
• A level playing field of independently audited Price/Performance benchmarks
• Forum open to challenge for 60 days by the entire competitive members
• Resource for many other types of benchmarks
• Resource for Performance Tuning Optimization

The best single resource from the TPC is probably the Complete Listing of TPC Results and includes up-to-date listings of all TPC/A, B, C, D, and E benchmarks detailing the company sponsor, system, database, operating system tested , and complete five-year costs including all

Page 1302

software and system maintenance support fees.

TPC/B: Current database stress test (not OLTP) detailing significant disk I/O, transaction integrity, all costs, moderate system and application execution time.

TPC/C: Simulated order entry of five complex, concurrent OLTP transactions; order delivery, payment record entry, status checks, and warehouse stock level environment test, including all costs, measured in Transactions Per Minute (TPM).

TPC/D: New decision support (DS) OLTP for large systems, processing 17 long running "real-world" queries against complex data structures.

TPC/E: New large-business "Enterprise"-database stress test for complex-query processing on very large databases.

TPC Quarterly Reports contain executive summaries of the TPC results published in the quarter as well as a complete historical benchmark listing. Executive summaries are two-page summations of the lengthy Full Disclosure Reports . The summations describe major components , performance data, detailed pricing of the system/configurations, and their related diagrams. Supplementary information included relates to competitive system and database analysis

topics.

Complete Full Disclosure Reports are an average of 100 pages but include detailed price breakdowns by components and the specific optimizations used to achieve the best performance results.

The TPC Price Performance ratio reports are available from within your organization, or you can telephone the TPC for reports and information directly in Sunnyvale, California, at (408) 295-8894.

Sample TPC Benchmarks

Following in Table 56.3 are a few sample TPC/A, TPC/C, and TPC/D benchmarks. The TPC consists of 44 members from the hardware and software industry, including

• Amdahl, AT&T/GIS(NCR), COMPAQ, Cray Research, DEC, Fujitsu, HP, Sequent, Silicon Graphics, Inc. (SGI), and Sun
• CA/Ingres, IBM, Informix, Microsoft, nCUBE, Novell, Oracle, SCO, and Sybase

TPC Results are subject to challenge by competitors and independently audited for 60 days prior to their publication. Benchmarks are summarized to include the complete five-year cost of the system including software with full support and administration costs. These figures are then expressed in total Transactions Per Second (TPS), (TPM), and the associated Price Performance Ratio.

Page 1303

Historical IBM 3090 Model 600J performance is 25 TPS, per processor. At six processors, this equates to 150 TPS/A v1.2. These machines cost $6 million to operate for five years , including software, maintenance, and support, thus generating a price performance ratio of $40,000.00@ Transaction. The figures shown in Table 56.3 reflect why the current trend is towards client/server and away from the mainframe.

Table 56.3. Sample TPC benchmarks.

Parallel Processing Platform Selection Criteria

Choosing your parallel processing platform requires specific selection criteria. Decisions you make regarding how you will evaluate the information and select the best current and future solution will have a long- term impact on your entire network user community.

When selecting a platform, consider your real-world experiences with your current file servers:

• Is your current query response time (TPS) and related performance fast enough?
• What is your current user community growth rate?
• How many concurrent and total users will you need to support in a few years' time?