Well the dust has finally settled and the Digital's
AlphaServer Systems have once again captured top honors as the
fastest UNIX Internet servers on the planet ! The benchmarks may
change (this time the Benchmark was SGI's own WebStone) but the
results remain the same. Internet AlphaServer systems remains
top dog in the race for Internet performance.
In a recent benchmark performance test using Netscape Communication Server V1.12 software, Digital's AlphaServer 1000 4/266 out performed a Sequent 8 processor Symmetry 5000 server, an HP 9000 Model D200 Web Server, a SUN Netra i20 (i600), and an SGI Webforce Challenge S on SGI's WebStone benchmark. In this case the Performance Graphs speak volumes. The Digital AlphaServer systems are the clear leaders in UNIX WebStone performance over all industry competitors, logging 343 connections per second at 18.35 Mb/sec maximum throughput. Overall the Internet AlphaServer 1000 out performed the competition by 3.7 - 8.6 times.If you still need more performance, check out the AlphaServer 2100A 5/300. This system with a single DECchip 21164 Alpha CPU managed 452 connections/sec and sustained throughput of 24.4 Mbits/sec.
Using 10 Mb Ethernet, the performance of the Internet AlphaServer 1000 4/266 was restricted by the network. Despite this, the AlphaServer 1000 still out performed all of the competition registering a peak of 129 connections/sec and throughput of 6.8 Mbits/sec to SGI's 55 connections/sec and 3.0 Mbits/sec. The Internet AlphaServer 1000 4/266 disposed of Sun, HP and Intel's platforms just as handily with both competitive systems managing between 40 and 66 connections/sec and 2 to 4Mb/sec of throughput.
Test Description:
Although Digital Equipment Corporation finds the
WebStone benchmark to be an inaccurate tool for measuring the
true performance of a Web server under real-world workloads,
the benchmark has been run for the purpose of providing comparative
data. Industry standard benchmarks like WEBperf (currently under
development), or Internet Load Benchmarks as seen in previous
"Performance Flash" reports are more accurate measures
of a system's performance potential. The WebStone benchmark, in
fact, is actually a relatively simple client-server benchmark,
developed by Silicon Graphics Inc., that measures the throughput
and latency of HTTP transfers. SGI offers the WebStone benchmark
software as freely available source code designed to measure simple
Web Server performance.
Test Results:
In addition to performance, the WebStone Benchmark
identified potential configuration weaknesses in competitive systems
by highlighting areas where bottlenecks could occur. Competitive
systems to the Internet AlphaServer 1000 4/266 began to saturate
disk, CPU, and memory subsystems long before a network saturation
occurred. As network traffic increases, their systems will be
unable to grow to accommodate the additional demand. The AlphaServer
1000 4/266 however, can sustain 6 to 9 times the network traffic
of SGI's or SUN's systems allowing for greater investment protection.
The addition of a 100 Mb Ethernet or FDDI networking card to the
SGI Webforce Challenge S, or the SUN Netra i20 would not significantly
improve their systems WebStones performance because the other
subsystems reach 100% utilization long before the network was
ever stressed. The initial benchmark was run using 10Mb Ethernet.
However, the AlphaServer 1000 4/266 approached the physical
limitations of the network long before CPU, disk and memory neared
100% utilization. By using FDDI, network bottlenecks were
removed and the additional headroom of the AlphaServer 1000 became
apparent. Using the WebStone benchmark the AlphaServer achieved
a maximum 343 connections/sec and throughput of 18.35 Mbits/sec.
Testing showed that AlphaServer 1000 4/266 saw a
2.7x performance increase for connections/sec and throughput,
after network bottlenecks were eliminated by replacing the network
. These results were even more dramatic for the AlphaServer 2100A,
registering a 3.4x improvement in both connections/sec and throughput.
Still Have Questions?
For more information contact your local Sales Office
or write to the Application Systems Engineering Performance Group
on-line at mail address asepg@zko.mts.dec.com.
Updated: Sunday, March 31, 1996