I have copied this from an email I subscribe to weekly.
Hello Everyone,
Today we am going to discuss hard drives and their interfaces. ATA: Advanced Technology Attachment. Formal name for IDE. This was the formal name adopted by American National Standards Institute (ANSI). Computer industry calls it IDE.
ATAPI: Advanced Technology Attachment Packet Interface. This is part of the EIDE interface standard and is also called ATA-2. This is the interface to connect non-hard drive devices to your computer. These devices include tape backup units, cd-roms, dvd-rom, and many others.
IDE: Integrated Drive Electronics. This is the standard interface between your motherboard and storage devices. IDE was developed to eliminate the need of a separate controller like in the days of MFM and RLL drives. I had one of these. IDE drives by name have the controller integrated onto the device. Look on the bottom of your hard drive. The circuit board is the controller. The cable connects to the interface. The standard was adopted by ANSI in 1990. It was based on the IBM PC 16-bit standard.
EIDE: Enhanced Integrated Drive Electronics. EIDE was developed for larger hard drives. Without EIDE you could not get above 528 MB on a hard drive. EIDE also incorporates DMA and support for multiple hard drives and non-hard drive devices called ATAPI devices. It uses a 28-bit Logical Block Address to allow up to 8.4 GB in size. A far cry from the IDE days of 512 MB and partitions of 32 MB in DOS 2.x. This allowed 16 Megabits per second transfer speeds. UDMA: Ultra DMA or Ultra DMA/33. This was the next generation in IDE technology. It incorporated 33.3 Megabits per second. UDMA was the first non-SCSI interface to use cyclic redundancy checking or CRC. CRC allows for error correction and resending of incorrect packets. UDMA supports Programmable Input/Output (PIO).
Mbps: Megabits per second
mode 0 3.3 Mbps
mode 1: 5.2 Mbps
mode 2: 8.3 Mbps
ATA standard allows
mode 3: 11.1 mbps
mode 4: 16.4 mbps
Ultra ATA/66 uses a 40-pin, 80-conductor wire to decrease cross talk and increase the interface speed to 66 Mbps. The additional 40 wires are grounds. Most Ultra ATA/66 hard drives are backwards compatible to the older interfaces. However, you will only get that interfaces performance. All the advances from the previous technologies carry through. The benefits are better than SCSI-UW performance under normal conditions. However, SCSI is still better for performance in servers and RAID level arrays.
Ultra ATA/100. Wow! 100 Mbps non-SCSI interface. The would surpass 80 mpbs in Ultra 2 SCSI. For a standard computer the performance would be fantastic. This is done by increasing the interface burst transfer rate. An old SCSI trick. The advances in noise suppression and better interface cables make these speeds possible. I am not sure if you will get 100 mbps. ATA/66 can get around 90% of promised speed due to some overhead. Even at this rate. A drive that could handle 90 mpbs would be great. Especially, at the low cost compared to Ultra 160 SCSI. Ultra 160 SCSI. I think you figured it out by now. As a true power user, I prefer SCSI. 160 MHz SCSI drives are the true rulers of the drive world. These come in 7200, 10K, and 15K RPM for maximum server level performance. It comes at a premium cost. Unlike IDE-level devices you need a controller. The Adaptec 29160 (single channel) and 39160 (dual channel) come at an added cost of $270 and above. The benefit is higher bandwidth and more devices. Each SCSI channel can handle 14 devices + controller. IDE can handle 4 devices. Devices are anything SCSI from hard drives, cd-roms, dvd-rom, scanners, tape backup units, etc. Your only limitation is cable length. The end devices on the SCSI chain are actively terminated to prevent cross talk. This is done on the cable internally and the last external device. The drives can outperform even ATA/100. I have seen them get 40-50 Megabytes per second on transfer tests under extreme load. SCSI uses busmastering to lower CPU usage. SCSI is the best solution for RAID.
RAID: Redundant Array of Independent Drives. Sounds like something from science fiction. It is how large servers protect their data, expand partitions to multiple drives, and improve performance. Raid comes in 3 main flavors: RAID 0, RAID 1 and RAID 5. Many others do exist. You can look them up on the ?net.
Lets build and imaginary array of 5 hard drives of 18 GB. You have 90 GB of potential storage. IDE RAID allows up to 4 devices.
RAID 0: Striping. This does not provide any fault tolerance to protect against drive failure. However it is the fastest way to configure. The drives are striped into 512 bytes sectors and are read across each drive instead of down the drive. Your data is read across all the drives.
RAID 1: Mirroring. This is the first fault tolerant system in RAID. You store the same information on two identical drives. If one drive fails the other drive immediately takes over. It does not use striping. It is slower due to the duplicate writes. Read performance is increased due to the fact you can read from either drive. This is the best way to protect a Microsoft Server?s boot and system partition. Duplexing is one step further. Mirroring shares a common controller. Controller failure will bring down a mirrored set. Duplexing brings a second controller for the second bank of drives. This increases fault tolerance. The bad point about this is you need more hard drives to do this. The cost of fault tolerance is increased by doubling your drives. You do not increase disk space. Drives are installed in identical pairs.
RAID 5: Stripe set with parity. Similar to RAID 1 with a unique twist. We have five hard drives @ 90 GB. We want to protect from a single hard drive crashing the entire array. Each stripe layer has 5 areas. Fault tolerance is brought in by using one of the striped areas to store parity information. The parity information is stored in the next location on each strip.
Here is 6 sector stripes. P=parity information and DÚta
1 2 3 4 5
1 P D D D D
2 D P D D D
3 D D P D D
4 D D D P D
5 D D D D P
6 P D D D D
These go on into the millions of 512 byte stripes
If drive 1 fails in this scenario, than this is what happens. You replace the hard drive and regenerate the stripe set.
Striped Rows 1 and 6: Parity is gone. Panic! Panic! Panic? No way. Since you have 5 drives and 4 data stripes intact. It recreates the parity information. Parity is restored.
Striped Rows 2,3,4 and 5. You have data missing. You have 3 data and 1 parity region. It takes the parity region and compares the remaining data regions. Basically and simplified: It subtract the known data from parity and replaces what was missing. This is a simple and non-proper way of looking at it. However, it gives you the idea of what is happening.
Backups are needed in case 2 drives fail. RAID 5 can only handle one drive failure. 2 or more drives requires the data be restored from backup. From this you reduce your data area by one hard drive. Our array would have 72 GB for data and 18 GB for parity. Not a waste due to the data protection. It is slower due to parity checks. I cannot stress the importance of tape backups.
Volume Set: NT and Win2K users probably have seen this term. You join mulitple hard drives into one large partition. It is not striped. It writes information in order of drives. In a multi-drive volume set the data is writen in order of the drives. First hard drive is filled then the second and so on. It just allows for a huge partition for large files. Drive failure requires you to restore from tape. No fault tolerance.
What is better IDE or SCSI RAID? Lets not talk cost. SCSI RAID is better for servers. SCSI RAID controllers have 32 MB+ of cached memory on-board and can handle upwards of 40+ hard drives. These controller can go into thousands of dollars. The cost for this performance is high. However, on a mission critical server, you do not look at cost. You want reliability and speed. For the average users on his home computer IDE RAID is a great way to get the benefit of RAID. I installed Win2K Server Ed to take full advantage of all RAID had to offer. WinNT and Win2K Server Editions can do software level RAID. It does not use a RAID controller. It is not as fast as a dedicated RAID controller. However, it does give you the benefits of RAID. Win9x, WinNT workstation, and Win2K professional do not have fault tolerance. With the release of Maxtor?s 80 GB monster drive I would not even look at RAID for the average user. If ATA/100 80 GB is not fast enough for you, you will want Ultra 160 SCSI.
The future is Infiniband. Infiniband is coming and will give you 2.5 GHz and up to 64,000 device capability. Search the web for more information on this technology. I hope this explains the difference hard drive interfaces.
Thisis from:
Michael A. McKenney
Electrical Engineer, CNE-IW & NW5, MCP
A+ Certified