Intel 170 Servers manual TCP/IP Function Megabit Ethernet LAN Virtual LAN, Network Operations

Here, a model 840 was subdivided into the partition sizes shown and a typical web serving load was used. A "hit" is one web page or one image. The kttpd is a kernel-based daemon available on Linux which serves only static web pages or images. It can be cascaded with ordinary Apache to provide dynamic content as well. The other is a standard Apache 1.3 installation. The 820 or 830 would be a bit less, by about 10 per cent, than the above numbers.

Network Operations

Here's some results using Virtual and 100 megabit ethernet. This pattern was repeated in several workloads using 820 and 840 processors:

TCP/IP Function	100 megabit Ethernet LAN	Virtual LAN
Transmit Data	50-90 megabits per second	200-400 megabits per second
Make Connections	200-3000 connections per	1100-9500 connections per
	second	second

The 825, 870, and 890 should produce slightly higher virtual data rates and nearly the same 100 megabit ethernet rates (since the latter is ultimately limited by hardware). The very high variance in the "make connections" relates, in part, to the fact that several workloads with different complexities were involved.

We also have more limited measurements on Gigabit ethernet showing about 450 megabits per second for some forms of data transmission. For planning purposes, a rough parity between gigabit and Virtual LAN should be assumed.

Gcc and High Optimization (gcc compiler option -O3)

The current gcc compiler is used for a great fraction of Linux applications and the Linux kernel. At this writing, the current gcc version is ordinarily 2.95, but this will change over time. This section applies regardless of the gcc version used. Note also that some things that appear to be different compilers (e.g. g++) are front-ends for other languages (e.g. C++) but use gcc for actual generation of code.

Generally speaking, RISC architectures have assumed that the final, production version of an application would be deployed at a high optimization. Therefore, it is important to specify the best optimization level (gcc option -O3) when compiling with gcc or any gcc derivatives. When debugging (-g), optimization is less important and even counterproductive, but for final distribution, optimization often has dramatic performance differences from the base case where optimization isn’t specified.

Programs can run twice as fast or even faster at high versus low optimization. It may be worthwhile to check and adjust Makefiles for performance critical, open source products. Likewise, if compilers other than gcc are deployed, they should be examined to see if their best optimizations are used.

One should check the man page (man gcc at a command line) to see if other optimizations are warranted. Some potentially useful optimizations are not automatically turned on because not all applications may safely use them.

The Gcc Compiler, Version 3

IBM i 6.1 Performance Capabilities Reference - January/April/October 2008
© Copyright IBM Corp. 2008	Chapter 13 - Linux	183