The PowerEdge 4600 Server: Performance Report
Ramesh Radhakrishnan, Ph.D (August 2002)
This article takes a comprehensive look at the performance of the DellTM PowerEdgeTM 4600 server on different benchmarks. The results of the benchmark tests demonstrate the performance capabilities of the server's unique features, which include dual Intel® XeonTM processors, PC1600 double data rate (DDR) memory, and a ServerWorksTM Grand ChampionTM High-End (GC-HE) chipset.
The DellTM PowerEdgeTM 4600 server replaces the Dell PowerEdge 4400 as the dual-processor departmental server. It provides enterprise-class performance for server applications and introduces major technological transitions for the departmental server platform.
Examining the features of the PowerEdge 4600
In a compact 6U chassis, the PowerEdge 4600 has dual Intel® XeonTM processors based on the Intel NetBurstTM architecture, a 400 MHz frontside bus (FSB), a ServerWorksTM Grand ChampionTM High-End (GC-HE) chipset, hot-plug PCI-X (Peripheral Component Interconnect Extended) slots, and 200 MHz double data rate memory (DDR200 or PC1600). Figure 1 compares the server's features with that of the PowerEdge 4400 server.
Figure 1. Comparing the features of the PowerEdge 4400 and the PowerEdge 4600 servers
The chipset used in the PowerEdge 4600 server helps to enable the server's high performance.
ServerWorks GC-HE architecture for Intel-based multiprocessor platforms. The GC-HE chipset delivers 3.2 GB/sec of memory bandwidth and up to 3.4 GB/sec of I/O throughput on a five-bus-segment architecture. It is also the industry's first platform to provide host PCI-X support.
The Intel Xeon processors and the technology they support are integral to the PowerEdge 4600 server's performance.
Dual Intel Xeon processors. Based on the NetBurst microarchitecture, these processors offer speeds starting at 1.8 GHz with 512 KB full-speed level 2 (L2) integrated cache and 400 MHz FSB.
Hyper-Threading functionality. This technology can improve performance of multithreaded code and multitasking operations through better use of processor execution resources.
Rapid Execution Engine. This feature allows execution of certain instructions at twice the processor's core frequency.
The PowerEdge 4600 server includes powerful memory capabilities for higher performance.
DDR SDRAM. DDR SDRAM doubles the data transfer rate to 1.6 GB/sec by transferring data on both the rising and falling edges of the clock.
DIMMs. Two memory cards offer combined support of up to 12 GB, four-way interleaved memory (dual in-line memory modules (DIMMs) installed in sets of four), which provides a theoretical peak memory bandwidth of 3.2 GB/sec.
Support for RAID and high-bandwidth data transfers enhance PowerEdge 4600 server performance.
RAID controllers. A PowerEdge Expandable RAID Controller 3, Dual Channel Integrated (PERC 3/Di), supports an end-to-end Ultra3/LVD (low-voltage differential) RAID solution.
Drive controllers. An integrated dual-channel Adaptec® AICTM -7899 Ultra3/LVD SCSI controller provides up to 160 MB/sec bandwidth and can be used with the PERC 3/Di for a RAID-on-Motherboard (ROMB) implementation.
Testing the PowerEdge 4600 performance
To discover the performance capabilities of the PowerEdge 4600 server, Dell tested it using SPECweb® 99, Microsoft® Exchange 2000 MAPI (Messaging Application Programming Interface) Messaging Benchmark 2 (MMB2), SPEC® CPU2000, and Ziff DavisTM NetBench® 7.0.2. The SPECweb99 benchmark measures Web server performance with demands on memory, network, and processor subsystems. Microsoft Exchange 2000 MMB2 simulates a messaging environment with demands on processor and I/O resources. SPEC CPU2000 is a compute-intensive benchmark that measures the processor, memory, and compiler performance of a computer system. NetBench 7.0.2 measures the performance of application servers in a client-server environment.
The SPECweb99 benchmark
SPECweb99 is a software benchmark developed by the Standard Performance Evaluation Corporation (SPEC), a nonprofit group of vendors, systems integrators, universities, research organizations, publishers, and consultants. SPECweb99 is designed to gauge a system's ability to act as a Web server for static and dynamic pages. The benchmark measures the maximum number of simultaneous connections that a Web server can sustain while still meeting specific throughput and error rate requirements.
In the SPECweb99 test, the PowerEdge 4600 outperformed all other two-way servers1 by sustaining 4460 connections running the Red Hat® Content Accelerator application and by sustaining 4320 connections running Microsoft Internet Information Services (IIS) 5.0 and the Scalable Web Cache (SWC) 3.0 application (see Figure 2 ). These results represent a significant performance improvement over the PowerEdge 4400 server with dual Intel Pentium® III Xeon processors at 800 MHz: 102.7 percent improvement running the Red Hat Content Accelerator under the Linux® operating system, and 307.5 percent improvement running IIS 5.0 and SWC 3.0 under the Microsoft Windows® 2000 operating system.
Figure 2. Top two-way servers in SPECweb99 benchmark, as of May 10, 2002
The results of this benchmark test also illustrate the performance and scalability of the PowerEdge 4600 server-it can deliver high levels of performance for memory-intensive, high-bandwidth applications. It is also the first Dell platform to use the ServerWorks GC-HE chipset, which provides high memory bandwidth and I/O throughput.
Microsoft Exchange 2000 MMB2
MMB2 is a benchmark from Microsoft that uses the Microsoft Exchange 2000 Server application to test mail server performance. The Microsoft LoadSim tool simulates user workloads on the system being tested, which runs the Exchange Server software. The MMB2 workload employs greater message sizes and more diverse actions than that of its predecessor, MMB (Medium Messaging Benchmark), which was used with versions of Exchange prior to Exchange 2000. MMB2 more accurately reflects the increased workload of messaging services deployed in today's demanding corporate e-mail environments.
Configured with dual Intel Xeon processors at 2.0 GHz, 4 GB of DDR memory, and two-way direct-attached SCSI storage, the PowerEdge 4600 server ran the Exchange 2000 Server Enterprise Edition application under the Windows 2000 Advanced Server operating system. The MMB2 test was performed using the LoadSim tool and the canonical MMB2 profile that represents a typical corporate e-mail user.
The results of the test showed that the PowerEdge 4600 server supported a benchmark load of 7,000 MMB2s (Figure 3 ). During four hours of steady test running, the server provided a weighted 95th percentile response time of 159 milliseconds (ms) for 7,000 Exchange (MAPI) users, with an overall processor utilization of 74 percent.
Figure 3. Exchange 2000 MMB2 results
In the Exchange Server MMB2 environment, the PowerEdge 4600 showed a 37.3 percent improvement over the PowerEdge 4400. The dramatic change in Exchange-related performance results from the high memory and I/O bandwidth provided by the ServerWorks GC-HE chipset and the processing power provided by the NetBurst microarchitecture of the Intel Xeon processors.
The SPEC CPU2000 benchmark
SPEC CPU2000 is a collection of industry-standard, CPU-intensive benchmarks developed by SPEC. The CPU2000 benchmarks are designed to compare compute-intensive performance across different hardware platforms. The benchmarks were developed from real user applications and measure performance of the processor, memory, and compiler on the tested system. Figures 4 and 5 show how the SPEC CPU2000 benchmark performance of the PowerEdge 4600 server compares to the performance of other PowerEdge servers. In these tests, the PowerEdge 4600 server was configured with one Intel Xeon processor at 2.0 GHz and 4 GB of DDR memory.
Figure 4. SPEC CPU2000 benchmark: Speed results
Figure 5. SPEC CPU2000 benchmark: Rate results
The SPEC CPU2000 benchmarks can be divided into two suites: CINT2000 for measuring compute-intensive integer performance and CFP2000 for measuring compute-intensive floating-point performance. The CINT2000 suite comprises 12 application-based benchmarks written in the C and C++ languages; the CFP2000 suite comprises 14 CPU-intensive benchmarks written in Fortran (77 and 90) and C.
The speed metrics, SPECint® 2000 and SPECfp® 2000, measure the ability of a computer to complete single tasks. The rate metrics, SPECint_rate2000 and SPECfp_rate2000, measure the throughput, or rate, of a machine carrying out a number of similar tasks. The rate metrics have traditionally been used to demonstrate the performance of multiprocessor systems.
When compared with the performance of the PowerEdge 4400 server, the PowerEdge 4600 server shows an impressive 55 percent improvement in the integer suite of the CPU2000 benchmark suite and up to 119 percent improvement in the floating-point suite, as shown in Figure 6 .
Figure 6. Performance improvement of the PowerEdge 4600 over the PowerEdge 4400 on the CPU2000 benchmark
The NetBench 7.0.2 benchmark
To test file-serving capabilities, the Ziff Davis NetBench 7.0.2 benchmark stresses and measures a server's throughput in megabits per second (Mbps). Eighteen different I/O requests represent the file-serving workload. NetBench uses a standard test suite to define the think times, ramp times, and number of clients and engines for each run. Clients represent physical machines; engines represent processes that execute multiple, simultaneous instances of the test suite on a physical client. The disk and network I/O subsystems are critical for obtaining optimum NetBench scores.
For the enterprise-level test suite of NetBench 7.0.2, a PowerEdge 4600 server with dual Intel Xeon processors at 2.2 GHz and 3 GB of RAM performed 28.4 percent faster than a PowerEdge 4400 server with dual Intel Pentium III Xeon processors at 1.0 GHz. Likewise, the peak throughput of the PowerEdge 4600 server was 38.5 percent higher than that of the PowerEdge 4400 server. Figure 7 shows the performance improvement that various configurations of the PowerEdge 4600 server achieved over the PowerEdge 4400 server. These configurations include three processor frequencies—1.8, 2.0, and 2.2 GHz—and three memory configurations—1, 2, and 3 GB.
Figure 7. NetBench 7.0.2 performance improvement achieved by various configurations of the PowerEdge 4600 over the PowerEdge 4400
Performance improvements over the PowerEdge 4400 server occurred for all configurations of the PowerEdge 4600 server, illustrating the superior performance capability of its chipset, processor frequency, and memory. Improvements in average performance ranged from 24.7 to 28.4 percent and improvements in peak performance ranged from 32.9 to 38.5 percent.
Upgrading to a more powerful departmental server
The PowerEdge 4600 server performed 25 to 300 percent better than its predecessor, the PowerEdge 4400 server, when tested by Web-server, mail-server, compute-intensive, and file-server benchmarks. Figure 8 summarizes the results of these benchmark tests.
Figure 8. PowerEdge 4600 performance summary
The powerful NetBurst-based Intel Xeon processors, faster FSB (400 MHz), DDR memory technology, and ServerWorks GC-HE chipset enable the Dell PowerEdge 4600 server to provide the compute horsepower required by server applications.
Ramesh Radhakrishnan, Ph.D. (firstname.lastname@example.org) is a design engineer consultant with the Dell System Performance and Analysis Lab. His responsibilities include performance analysis of Dell servers and characterization of enterprise-level benchmarks. Ramesh received a Ph.D. in Computer Engineering from the University of Texas at Austin.
FOR MORE INFORMATION
Dell PowerEdge 4600 benchmark studies
SPECweb99: http://www.dell.com/downloads/us/ pedge/pwrnote_4600_specweb.doc
SPEC CPU2000: http://www.dell.com/downloads/us/pedge/pwrnote_4600_cpu2000.doc
Benchmark Web sites
Microsoft Exchange 2000 MMB2: http://www.microsoft.com/exchange/techinfo/ planning/2000/PerfScal.asp
Microsoft LoadSim: http://www.microsoft.com/exchange/downloads/2000/LoadSim.asp
Standard Performance Evaluation Corporation (SPEC): http://www.spec.org
SPEC CPU2000: http://www.spec.org/osg/cpu2000