Bitzenbytes.com

Hey NoClue…why do you always go on about SCSI? Do you have ‘scuzzyitis’?

Yeah…I guess I do.  SCSI rulez!    In this post, I’ll try to explain why.

What is SCSI?

quote:

---

SCSI, or Small Computer System Interface is a “high speed, intelligent peripheral I/O bus with a device independent protocol” (from the Paralan SCSI FAQ)

---

Huh?

Okay…it’s an I/O bus (2-way pathway for data) that is high speed (currently up to 160MB/sec, but 320MB/sec isn’t far off), intelligent (it uses things like disconnect/reconnect and it can pass data along the bus without pestering the CPU) and device independent (it doesn’t care what the device is, a hard drive, a scanner, a printer, etc).

The SCSI bus consists of (at least):

• A Host Adapter – Your SCSI card or onboard chip. Also called Host Bus Adapter (HBA) or SCSI Controller. I’ll usually refer to it as SCSI card in this post.
• A SCSI device – this can be a hard drive, CD-ROM, CD-R/RW, DVD-ROM, optical drive, tape drive, scanner, printer, etc.
• Cabling – this can be internal or external…or both.
• Terminators – these can be active or passive, they can be built into the device or plugged into the bus. There must be 2 terminators on each bus…no more, no less.
  

When dealing with SCSI devices, you’re likely to hear terms like Fast SCSI, Ultra SCSI, Ultra2 SCSI, Wide SCSI, etc. These various terms describe the amount of data that can be passed through the SCSI bus. We’ll get into further detail later.

Why do I need to ‘go SCSI’? Isn’t IDE/ATA okay?

Well…you don’t necessarily. IDE/ATA is fine for many applications, especially on a machine with a single hard drive.

Let’s look at some of the ‘Pros & Cons’ of both:

IDE/ATA

Pros:

• IDE/ATA is less expensive than SCSI. There is less development and testing required than with a SCSI device, and therefore a much higher volume of IDE/ATA devices are produced.
• IDE/ATA is supported by the BIOSes on most motherboards. In most cases, the controller is built-in.
• There is less overhead per command with IDE/ATA.
  

Cons:

• You are limited to 4 devices on most systems; or at most 8, with an additional controller card.
• IDE/ATA devices are limited to disk drives, with some support for tape.
• IDE/ATA is single threaded. In other words, commands cannot overlap, even if you have a second drive.
• IDE/ATA has limited support for DMA (Direct Memory Access – where the drives communicate directly with memory, bypassing the CPU). For those systems without DMA, the CPU is involved in the transfer of all data, which prevents it from doing other tasks.
• With IDE/ATA and ATAPI there can sometimes be compatibility issues. There are drives that will work fine when installed alone, but won’t work when another drive is added. These issues are rare, but they do happen.
• IDE/ATA devices do not handle scatter/gather operations very well.
  

 
SCSI:

Pros:

• Newer SCSI systems support up to 15 devices per bus (16 counting the controller)
• SCSI devices can be disk drives, tape drives, printers, scanners…almost any peripheral device.
• With SCSI, all commands can overlap with commands on other devices.
• SCSI usually uses DMA to transfer data, thus freeing the CPU for other tasks.
• The largest, fastest devices come out as SCSI devices before IDE/ATA.
• Most SCSI cards can do scatter/gather DMA.
  

Cons:

• SCSI devices usually require more complex firmware and testing, and offer higher performance. Thus SCSI is usually more expensive than IDE/ATA.
• SCSI is slightly more complicated to install than IDE/ATA.
  

Let’s look at some of these points in further detail…

Expense:

SCSI is more expensive than IDE/ATA…no doubt about it. You’ll need to decide if the extra performance is worth the extra expense.

If you plan to do data intensive (either CPU-intensive or storage-intensive) things like digital video editing, you really don’t have any choice; SCSI is the only way to go.

Device Limitations:

Most IDE/ATA systems have 2 channels (Primary and Secondary), with up to 2 devices per channel (Master and Slave), which offers you a maximum of 4 devices. These devices must be internal, and are limited to drives (hard disks, CD-ROM/R/RW or DVD, ZIP/LS120, and a few tape drives).

If you add a separate controller (such as an ATA-100 controller), you get an additional channel, and 2 more devices.

SCSI can offer you up to 16 devices (the SCSI controller card counts as a device, so it’s really 15 and a controller). Even Narrow SCSI offered you 8 (or 7 and a controller). These devices can be internal or external and can be not only all of the drives previously mentioned, but also printers, scanners, optical and magneto-optical drives, etc.

Overlapping Commands and Connect/Disconnect:

IDE/ATA is a single-threaded architecture. This means that of the 2 drives on a given channel, one will always be idle while the other is being read or written to. SCSI allows devices to connect/disconnect at will, thereby sharing the bus more efficiently. In other words, while one device is seeking or rewinding, another device can transfer its data.

IDE/ATA will not allow more than one I/O request per channel at a time, even to different drives. SCSI allows up to 256 commands to be queued, and they can be completed out of order. To quote from the Official comp.periphs.scsi FAQ at Gary Field’s SCSI Info Central:

quote:

---

For instance, process 'A' needs to read a block. The request is sent to the drive, the disk head starts to move, and process 'A' blocks waiting for it. Then, process 'B' is allowed to run; it also reads a block from the disk. Process B's block may be sitting in a RAM cache on the SCSI controller, or on the drive itself. Or the block may be closer to the head than process A's block, or on a different drive on the same cable. SCSI allows process B's request to be completed ahead of process A's, which means that process B can be running sooner, so that the most expensive chip - the system CPU - tends to spend less time twiddling its thumbs. Under ATA, the process B request cannot even be sent to the drive until the process A request is complete. These SCSI capabilities are very valuable in a true multi-tasking environment, especialy important in a busy file server, and useless under DOS, which cannot take advantage of them.

I tend to hear from people, 'Well, I never use multitasking' because they never actively run two programs at once - all but one are 'just sitting there'. Consider what happens though, when you minimize a window which uncovers parts of four other application windows. Each of those applications is sent a message telling it to update part of its window; under win95, they will all run concurrently to perform the update. If they need to access disk (usually because of virtual memory) the smoothness of the update can depend a lot on the disk system's ability to respond to multiple independent read requests and finish them all as quickly as possible; SCSI is better at this.

---

 
DMA and Scatter/Gather:

In modern operating systems (Windows, Unix, Linux, Netware, etc…pretty much everything but DOS and Win3.x) the CPU can go off and do other processing while data is transferred directly into memory using DMA. These operating systems also use what is known as virtual memory…a space on your hard drive that the system uses as memory. In virtual memory systems, a single 64KB chunk of data can be spread over up to 16 memory pages. The term scatter/gather refers to what happens to the data during the DMA operation. When data is read from a device, the sequence of bytes are scattered onto the various memory pages. When data is written to the device, the data is gathered from the various memory pages and organized into a sequence of bytes to be transferred to the device.

A good SCSI card, given a single request, will handle this scatter/gather operation all by itself. IDE/ATA must interrupt the CPU to handle this.

 
Ease of Installation:

IDE is slightly easier to install, the controller is usually built into the motherboard, and all that is required is setting some jumpers for Master or Slave. SCSI requires installing a controller card (usually, some mobos do come with built-in SCSI), setting jumpers for the SCSI ID, and adding terminators…but as we’ll see, this stuff isn’t very difficult either.

Okay…any more questions?

Yeah…what’s all this stuff about Wide/Narrow SCSI…or Ultra/Ultra2 SCSI…what’s it all mean?

Glad you asked.

As the SCSI standard has evolved, the busses have gotten faster and wider. They’ve gone from the old SCSI-1 at 8-bits wide and 5MB/sec to the coming-soon-to-a-vendor-near-you Ultra320 SCSI at 16-bits wide and 320MB/sec.

The following table may help:

 
As you can see in the table, Narrow SCSI means 8-bits wide and support of up to 8 devices; Wide SCSI means 16-bits wide and support of up to 16 devices.

Narrow SCSI is said to contain the low byte, where Wide SCSI contains both the low byte and the high byte. This will become more important as we discuss termination.

SCSI devices can be single-ended or differential. Single-ended means that there is one wire for each signal that needs to be sent across the bus. Differential means that there are two wires for each signal; the first wire carries the signal, the second wire carries its logical inversion. The receiving device takes the difference between the two (thus the name differential). Differential SCSI is less susceptible to noise, allowing the bus to be longer.

Differential devices are further divided into HVD (High Voltage Differential) and LVD (Low Voltage Differential). Before the new LVD interface was designed, HVD was just called Differential, but the name has been changed to distinguish it from LVD.

HVD devices could not exist on the same bus as single-ended (SE) devices. LVD devices can switch into SE mode and co-exist with them. LVD also offers a high speed bus up to 12 meters in length. LVD devices have no on-board termination, so a separate terminator is required.

I would be remiss if I didn’t discuss the SCSI standards at this point (Thanks Bob ). Unfortunately this area is pretty fuzzy, especially when dealing with vendors/sales staff. But, here goes…

SCSI-1

SCSI-1 is a complete document that includes standards for all the physical and protocol layers. The connectors used for SCSI-1 were the low-density 50-pin internal and external connectors; the external connector is also called the Centronics connector. SCSI-1 is now obsolete, according to the SCSI Trade Association.

SCSI-2

Way back in 1985, when the SCSI (now SCSI-1) standard was still being hammered out, a group started working on an extension of that standard, which became SCSI-2. SCSI-1, unfortunately, did not bring about ‘common compatibility’ between manufacturers…different manufacturers added different undocumented ‘features’ to their devices. SCSI-2 addressed this problem while still assuring backward compatibility for devices that were designed to the SCSI-1 spec. SCSI-2 is also a complete document.

For more information about the differences between SCSI-1 and –2, check this out.

SCSI-3

Where SCSI-1 and –2 were complete documents, SCSI-3 is a collection of documents that cover the physical, protocol and command set layers. In most of these separate documents, the –3 has been dropped…each document has it’s own revision number.

For more information, see this.

The tricky part about the standards is that they have always lagged behind the real world. Why did SCSI-1 end and SCSI-2 begin? Every once in a while, you’ve got to stop and get a standard published.

If you stick to the terms endorsed by the SCSI Trade Association when dealing with vendors, you have a better chance of getting what you want.

SCSI is also different than IDE/ATA in that it requires each end of the bus be terminated.

What do you mean...terminated?

SCSI cables are basically transmission lines. Signals are bounced off of the ends and reflected back to devices on the bus. Termination merely stops those signals from bouncing back.

The key to termination is to remember that for each SCSI bus, there must be 2 terminators and there can only be 2 terminators…one at each end of the bus. Drive letters, SCSI ID’s, etc. don’t matter…the terminators must be at the physical ends of the SCSI bus.

Let’s look at some examples:

Probably the simplest SCSI bus (shown above) is a single device (an internal hard drive) connected to the SCSI card. In this example, both the card and the drive must be terminated.

In this example, we have added an external device, a scanner. In other words, we have moved the end of the bus from the card to the scanner. We will also need to move the termination to the new end of the bus, so we will un-terminate the card and terminate the scanner.

Now we have added 2 more hard drives and a printer. Again, we need to move the termination to the physical ends of the bus. We will un-terminate the original hard drive and the scanner, and terminate the new drive and the printer.

This can sometimes be confusing when compared to IDE/ATA. When setting up an IDE/ATA drive, its location on the cable is unimportant…the Master/Slave settings are set using jumpers on the drive.

When setting up a SCSI drive, you need to pay attention to cable locations, so you’ll know where the bus should be terminated.

SCSI ID’s, on the other hand, are set with jumpers on the drive, and like IDE/ATA, location on the cable is unimportant (more on SCSI ID’s later).

There is also some confusion about the way terminators effect the SCSI bus. I have heard people say that they had added a device after the terminator, and were surprised that the SCSI controller could still “see” the device. The effects of incorrect termination are unpredictable, but you will normally “see” the device.

Termination can be tricky when dealing with wide vs. narrow devices (in fact, connecting narrow devices to a wide bus can be tricky for a number of reasons). As we discussed before, wide devices contain both the low byte and the high byte, where narrow devices contain only the low byte. If you connect the narrow device to the end of your wide bus (using a suitable adapter) and terminate it, the high byte will be left un-terminated. In this case, you’ll need an adapter with high byte termination.

Termination Methods.

There are two types of terminators, Active and Passive.

Passive Terminators are a group of resistors that dampen reflected signals from the ends of the bus. They rely on the TERMPWR (+5v) line from the cable for their power.

Active Terminators add a voltage regulator rather than relying on just the TERMPWR line (which may or may not be exactly +5v).

If you are buying terminators, buy Active terminators. Anything other than single-ended devices require them.

There are many different ways to terminate devices…jumper settings, software settings, auto-termination, and plugging/unplugging terminators into/out of the individual devices.

I usually try to get separate internal and external active terminators and place them at the ends of the bus, disabling any on-board terminators or auto-termination. That way, any internal devices can just be plugged into the cable, and if I add any external devices, I just move the terminator.