Before you start
Objectives: Learn about the advantages and disadvantages of common storage interfaces in server environments.
Prerequisites: you have to know all about typical storage devices and their interfaces.
Key terms: disk, SCSI, hard, SATA, drive, storage, PATA, system, server, controller, master
Hot Swappable Hard Drives
The first thing we should ensure is that our server supports hot swappable drives. This feature allow us to remove the hard drive from the computer while the system is running. We power off the hard drive itself, then remove it from the system, and plug a new drive back in. If we use hot swap, the system will automatically recognize the new hard drive and configure it. Hot swap is usually implemented by using SCSI drives, or newer SATA drives. Typically, most server machines will support this feature.
In addition to hot swap, some servers will support Cold Plug feature. With cold plug we can also power off our drive and replace it with a new one while the system is running. The difference is that with cold plug the system won’t automatically recognize and configure the new hard drive. With cold plug we have to manually tell the system that we have a new drive in our computer.
Types of Storage Interfaces
As you should already know, for directly attached storage we basically have three types of interfaces: PATA, SCSI and SATA. Another option is to set up a fibre channel SAN.
Parallel ATA (PATA) hard drives, often called IDE, EIDE and ATAPI (ATA evolved from IDE interface), are relatively cheap to implement, the speed is OK, and the capacity can be large. ATA is the standard which was used for transferring data from hard disks and CD-ROM drives, but it is superseded by SATA. The problem with PATA drive is that its disk controller is implemented on the hard drive itself. That fact makes those kind of hard disks less suitable for server computers. When we configure PATA drives, one drive is set to be the “master” disk, and another one is set to be the “slave” disk. When we do that, the disk controller on the master disk actually controls the slave disk as well (the controller on the slave disk is disabled). If the controller on the master disk malfunctions, the data on both the master and the slave disk could get corrupted. This is why this kind of configuration is not suited for software RAID mirroring. If we set up RADI 1 (mirroring) between a master disk and a slave disk, we could lose data on both disks if we have a problem with master disk. In fact, mirroring feature will simply mirror all errors from one disk to the other disk. Another disadvantage of PATA is that we are limited to a number of 4 hard drives (two channels) on a typical motherboard. Because of all those problems, it is not recommended to use PATA disks on server machines. Also, the maximum speed of the disks is 133 MB/s, which is great for desktop computers, but for servers we should use faster interfaces.
The interface which is often used on servers is Small Computer System Interface (SCSI). The first advantage of SCSI when compared to PATA is the speed. With SCSI we can get a 640 MB/s (Ultra-640 SCSI). We can also install more drives in our system. With SCSI we can have 15 hard drives on our system. SCSI is also great for mirroring and duplexing. With mirroring we use one SCSI controller and with duplexing we use two SCSI controllers. With duplexing we basically eliminate a single point of failure which is the SCSI controller itself. In contrast to PATA drives, SCSI drives will typically last longer. The downside of SCSI is the higher cost when compared to PATA, and they are a bit harder to configure (devices must have a device ID and the bus must be terminated).
The newest storage interface which can be used on servers is Serial ATA (SATA). The advantage of SATA disks is that they are relatively cheap. Also, the speed of SATA disks is similar to SCSI, which means that SATA is also very fast. At the moment of writing this article the last revision of SATA is SATA revision 3.0 with the speed of 600 MB/s. SATA also works great for duplexing and mirroring. The configuration of disks is very easy (we simply connect the disk). SATA also provides built-in support for disk protection.
For storage on server machines we can also use a Fiber Channel SAN. This is probably the best storage solution for servers. In this configuration we connect our server to the storage device, or multiple storage devices, over network by using fiber optic cabling. Fiber channel SAN provides high speed, external storage which can be located on different locations, and we get a support for clustering if we need to implement high availability. The downside of fiber channel SAN is the cost.