1.1 Performance Tuning Examples

Several places in this book cite real performance numbers that I measured on a test network. It is appropriate that some information be provided up front about the system and test methodology used in arriving at the data presented in this book.

The systems used to generate the data found in this book are not configurations that I would recommend using in actual practice. This discrepancy is intentional. I selected these particular configurations as test platforms for the following reasons:

  1. It is important to understand how performance can be affected in both CPU-bound and I/O-bound server situations. Therefore, I use server configurations that have been artificially manipulated to be CPU bound and I/O bound. In fact, these two configurations will be referred to as "CPU-bound" and "I/O-bound" servers in this book.

  2. The extent to which the tuning advice provided in this book affects different servers at different sites will vary widely. For example, changing the delivery mode employed by sendmail will affect a two-processor Sun Enterprise 220R with a Sun Netra st A1000 disk storage array differently than it affects a 2.0GHz Linux server using software RAID to stripe four disks together. It would be improper to make assumptions about the effects on a site's performance based on data gathered under a different configuration. Hopefully, the temptation to do so will be reduced if data are presented regarding server configurations that no one would use.

  3. It is easier to be confident in results obtained using a small number of machines as part of a test network. The fewer and simpler the machines involved, the less likely that someone would forget to go through all the tasks necessary to ensure that the conditions have been repeated exactly from one test run to the next. To make the testing easier by making the load generation facilities simpler, the bottlenecks in the test systems cited in this book have been exaggerated. Of course, in "real life," we don't always have a lot of choice in how complex an environment we want to test. Because this book will be read by many people, however, it is especially important to get the test results correct here.

  4. With whatever environment is being tested, it is important to ensure that sufficient load generation capacity and network bandwidth, not to mention adequate power and cooling, are available to perform accurate tests. At the time of this writing, I don't have unlimited access to a high-performance test lab. Nevertheless, interesting results can be obtained from just about any available hardware.

For the purposes of this book, we're primarily interested in whether a given change will affect system performance positively or negatively. We're also interested in how large an effect we might expect each change to have on our system, although we need to be skeptical that precise numbers generated in one environment will apply in another environment.

Now that the disclaimers have been made, let's look at the two test servers. The CPU-bound server has the following configuration:

  • The server is a Sun Sparc 5 (not Ultra 5, Sparc 5) clone, vintage circa 1996.

  • It has a single 85MHz MicroSPARC II processor.

  • It has 32MB of RAM.

  • It's running Solaris 2.6 with some of the jumbo patches installed. No attempt has been made to tune the kernel.

  • It contains two 2GB Seagate Hawk (5400 RPM) disks on a SCSI-2 bus.

  • It has a single 10 Mbps Ethernet card to connect it to the test network.

The I/O-bound server has the following configuration:

  • The server is a contemporary PC-based server.

  • It has a single 1.6GHz Intel Pentium 4 processor.

  • It has 256MB of RAM.

  • It's running the Linux 2.4 kernel specifically, Red Hat 7.2. The operating system is unpatched.

  • It has a 4GB Seagate Hawk (5400 RPM) disk for the operating system and home directories. A second 2GB Seagate Hawk (also 5400 RPM) is used for testing. It has been split into two 1GB partitions. The inner (slower) partition is used for all testing. These are SCSI-2 disks.

  • Its mainboard provides a single 100 Mbps Ethernet connector to connect it to the test network.



sendmail Performance Tuning
sendmail Performance Tuning
ISBN: 0321115708
EAN: 2147483647
Year: 2005
Pages: 67

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net