Using High-performance I/O Subsystems

From InterBase

Go Up to Hardware Configuration


A multiuser database server’s hard drives are no place to be thrifty, especially in today’s market of inexpensive storage. Configuring a relatively high-end I/O system is a cost-effective way to increase performance.

Slow disk subsystems are often the weak link in an otherwise high-performance server machine. The top-rated CPU and maximum memory helps. But if a cheap disk I/O interface limits the data transfer rate, then the money spent on the expensive components is wasted.

It’s not appropriate for this document to recommend a particular configuration. The technology changes so quickly that any recommendation here would be outdated. When you specify the machine for a server platform, research the best hardware solution available.

Read the following guidelines for principles:

  • Advanced SCSI technology offers superior I/O throughput. The following graph illustrates the relative maximum throughput of different disk interfaces.
  • The external interface capacity usually exceeds the internal or sustained transfer rate of any individual device. Only systems that use multiple disk devices make full use of a high-capacity I/O interface.
  • Bus-mastering I/O controllers use less CPU resources. This is particularly important on I/O-intensive server machines. SCSI is generally bus-mastering, and newer PCI EIDE interfaces are bus-mastering. IDE is not.
  • Use a disk controller with built in cache memory. The controller cache reduces the need for the operating system to use system RAM for disk cache.
  • Don’t assume all disks of a given size perform equally; research performance ratings made by independent testing labs.

Advance To: