Managing RAID on Linux - Derek Vadala [28]
Likewise, there are quite a number of reports about using autotermination of SCSI chains. Autotermination is built into controllers and disks. If you experience problems, you may wish to manually disable autotermination (which is a controller BIOS setting or a jumper on your hard disk) and actively terminate the chain at both ends. You can purchase terminators from the vendor who provides your SCSI cables, or even from a local computer store.
Growth
SCSI certainly has a much better upgrade path than ATA. Device-per-channel limits make SCSI much easier to deal with when you need to increase the size of an array. If you're on a budget, you might find it worthwhile to purchase an expensive SCSI controller, along with drives that are one or two technologies behind the current trends. Buying the latest and greatest SCSI card will increase the final price of a system by only a few hundred dollars. Buying the most cutting-edge disks, on the other hand, will affect system price by a few hundred dollars per drive. So while you can save costs by purchasing older drives initially, you won't have to discard your SCSI controller if you decide to upgrade to faster drives a few months later. Scaling back the original drive purchase initially might even place a hardware RAID controller within your budget.
Summary of SCSI versus ATA
SCSI supports more devices than ATA (although configuring many devices can be a challenge for many users and administrators). ATA is in more widespread use than SCSI, and that might make it easier to get hold of enough hardware to build a decent array. Some naysayers argue that SCSI is more confusing than ATA because SCSI users are faced with termination and drive placement considerations. Others are quick to point out the autodetection and block addressing problems with which ATA users must contend.
ATA can access only one device at a time, meaning that the benefits of parallel I/O under RAID are wasted. SCSI can address multiple devices concurrently and does not require the CPU to manage I/O, leaving more processing power for users and applications.
Table 2-6 summarizes the differences between ATA and SCSI. I think you will find that ATA is a cheap and usable way to quickly build arrays for both desktops and low-usage production systems, but that SCSI is the best choice for large systems and applications that require extremely intense I/O.
Table 2-6. The differences between SCSI and ATA
Feature
SCSI
ATA
Device limit
7 or 15 per channel
2 per channel
Maximum cable length
12 meters
~.5 meters
External devices
Yes
Not without special hardware
Termination required
Yes
No
Device ID
Yes
No (master/slave only)
Extra CPU load
No
Yes
Concurrent device access
Yes
No
Cost/availability
Expensive; need to add on
Cheap; built into most motherboards
I have excluded data throughput differences between SCSI and ATA from Table 2-6 because throughput with each protocol is typically limited by disk rather than by channel. Both SCSI and ATA will perform roughly equally in single-disk operations (assuming that similar specifications are compared). That being the case, SCSI supports many drives per channel, whereas ATA supports only one, from a usability standpoint. Thus, with SCSI, it's a lot easier to use the bandwidth you have available, while with ATA, it's really not possible.
Other Disk Access Protocols
Because RAID is oblivious to the hardware and disk architecture on which it is built, you can use any disk protocol that the Linux kernel supports to build an array. Indeed, if a newer, faster, and more reliable disk protocol (such as Serial ATA) were released this year or the next, it would only increase the usefulness of RAID. Furthermore, if a breakthrough in solid-state media happened in the next few years, these devices could also be grouped into arrays. While disk capacities and throughputs continue to increase, they nonetheless continue to fall behind the curve of increasing user needs.
Choosing Hard Drives
Hard drives represent