UNIX System Administration Handbook - Evi Nemeth [446]
24.10 TOOLS
A well-outfitted system administrator is an effective system administrator. Having a dedicated tool box is an important key to minimizing downtime in the event of an emergency. Table 24.1 lists some items you should probably keep in your tool box, or at least within easy reach.
Table 24.1 A system administrator’s tool box
a. And maintenance contract numbers if applicable.
1. That is, unless memory is part of a package deal; some of these deals are OK.
2. Power-cycling is the process of turning a machine off, waiting 30 to 60 seconds for the capacitors to drain, and then turning the machine back on again.
25 Performance Analysis
Performance is one of the most visible characteristics of any system, and it’s often high on the list of complaints from users. Many users are convinced that their computers could run twice as fast if only the administrator knew how to properly tune the system to release its vast, untapped potential. In reality, this is almost never true.
One common fantasy involves tweaking the kernel variables that control the paging system and the buffer pools. Once upon a time, there were situations in which this was necessary and prudent. These days, it is usually a bad idea. The most likely result is that you will reduce your system’s overall performance and not even be aware of what you’ve done, all the while congratulating yourself on being such a clever kernel hacker.
Modern kernels are pretuned to achieve reasonable (though admittedly, not optimal) performance under a variety of load conditions. If you try to optimize the system based on one particular measure of performance, the chances are high that you will distort the system’s behavior relative to other performance metrics and load conditions. It seems easy to get results, but the gains are usually illusory.
In particular, take everything you read on the web with a tablespoon of salt. In the area of system performance, you will see superficially convincing arguments on all sorts of topics. However, most of the proponents of these theories do not have the knowledge, discipline, and time required to design valid experiments. Popular support means absolutely nothing; for every hare-brained proposal, you can expect to see a Greek chorus of, “I increased the size of my buffer cache by a factor of ten just like Joe said, and my system feels MUCH, MUCH faster!!!” Right.
System performance is not entirely out of your control. It’s just that the road to good performance is not paved with magic fixes and romantic kernel patches. The basic rules are these:
• Don’t overload your systems or your network. UNIX gives each process an illusion of infinite resources. But once 100% of the system’s resources are in use, UNIX has to work very hard to maintain that illusion, delaying processes and often consuming a sizable fraction of the resources itself.
• Collect and review historical information about your system. If it was performing fine a week ago, an examination of the aspects of the system that have recently changed is likely to lead you to a smoking gun. Keep regular baselines in your hip pocket to pull out in an emergency.
This chapter focuses on the performance of systems that are used as servers. Desktop systems typically do not experience the same types of performance issues that servers do, and the answer to the question of how to improve performance on a desktop machine is almost always “Upgrade the hardware.” Users like this answer, because it means they get fancy new systems on their desk more often.
25.1 WHAT YOU CAN DO TO IMPROVE PERFORMANCE
Here are some specific things can you do