Talk:SCSI

Fiber Channel is not an alternative spelling of Fibre Channel.

SPI vs SCSI

The concepts relevant only to SCSI Parallel Interface should be separated from the article on SCSI proper. Although SCSI-1 and SCSI-2 include SPI as a central part of the protocol, SCSI-3 has completely split the framework into separate layers. --Cy jvb 20:20, 2 February 2006 (UTC)[reply]

This was a good suggestion so I have split out the SPI content and created a new page called Parallel SCSI.
Neilm 12:43, 7 February 2006 (UTC)[reply]

Pleonasm?

"SCSI is pronounced "scuzzy" when spoken aloud"... Isn't this a bit pleonastic? --Edcolins 11:51, Oct 28, 2004 (UTC)

"while occasional attempts to promulgate the more flattering pronunciation "sexy" have never succeeded." Nowadays we're trying to get it to be pronounced "sucksy" ;) --148.84.19.92 15:25, 9 Mar 2005 (UTC)

IDE?

The article says that PATAPI is SCSI over IDE. Isn't this a misnomer? As I recall, parallel ATA is not the only form of IDE, though PATAPI is, of course, over parallel ata.

Not quite; something the article doesn't make clear is that SCSI is a command set as much as (and, now even more than) it is an interface. If you look at the official specs for ATAPI, it is indeed sending SCSI commands over the ATA bus; in fact, most ATAPI devices these days are actually SCSI devices that speak enough ATA to get by. -lee 07:11, 13 Jun 2005 (UTC)

spice up SCSI

good article, Mac users in particular grew up on SCSI and SCSI peripherals

How about some pics of the typical fat cables and giant terminators that people remember:

http://images.google.com/images?q=scsi&hl=en&lr=&sa=N&tab=wi

Also what about the SCSI logo ?

http://images.google.com/images?hl=en&lr=&q=scsi+logo&btnG=Search

I agree that this article needs some pictures. Anyone willing to take some? 203.208.80.13 00:51, 14 September 2005 (UTC)[reply]

I've added some self-made images including the SCSI logo, a SCSI terminator, 25 pin SCSI cable, and Mac SCSI port. Feel free to move them around in the article if you feel some of them might be better placed. --CB1226 23:37, 2 March 2006 (UTC)[reply]

SCSI protocol

This article only seems to talk about devices...as far as I know, in the Linux kernel (for 2.6 at least), firewire IEEE1394 / USB storage devices / Ipod's, etc. they also use the SCSI protocol to communicate to the kernel. Any elaboration on this? -- Natalinasmpf 02:27, 19 Apr 2005 (UTC)

You are correct; most of this article focuses on what is now officially referred to as the SCSI Parallel Interface, which is just one part of SCSI-3 and has little to do now with the SCSI command set (which is, as you and others have noted, is implemented on things that are about as far from SPI as you can get). At some point (not tonight, since it's late) I'll go through and sort things out. -lee 07:05, 13 Jun 2005 (UTC)

I have created a new page called Parallel SCSI and moved some content to it. Neilm 12:45, 7 February 2006 (UTC)[reply]

June 22, 2005:

The article should mention "packetized" SCSI that was introduced with Ultra-320. IBM preferred the term "information units" over "packets". I think this was around 2002 or early 2003.

Also, the "Ultra" designations were promoted by the SCSI Trade Association but not used in the ANSI standards. After SCSI-2, the physical interface standard was separated from the logical layers, and became SPI (for SCSI Parallel Interface) and went through generations SPI-1, SPI-2, SPI-3, SPI-4 and SPI-5. STA wanted marketing-oriented nomenclature that reflected the actual maximum data transfer rate of the bus. Note that the SPI acronym can easily cause confusion with other standards.

I don't think "All SCSI standards have been modular, defining various capabilities which manufacturers can include or not" is true of SCSI-1 or SCSI-2, which were monolithic specifications, unless "modular" means something different from what I take it to mean.

I didn't write that part. But certainly SCSI-2 has wide scope for variation in device capabilities, with all sorts of contexts in which initiator and target are supposed to negotiate for a set of common capabilities.

I think this is too opinionated for the article, but to my mind one of the weaknesses of SCSI is that it is a humongous specification--SCSI-2 is about an inch thick--and it is very easy for devices to fail to implement some portions of it correctly. Obviously, if an Adaptec card doesn't work properly with Seagate disks they'll catch it and fix it, but obscure corners of the specification can be ignored, resulting in SCSI cards that work with most but not all devices, and vice versa.

It is similar to problems encountered with TIFF in, say, the early nineties, when it was quite possible to have a TIFF file that completely conformed to the TIFF spec but which could not be opened by PageMaker.

Another problem, to my way of thinking, is that the safety margins on things like cable length appear not to be conservative enough. I notice that Adaptec consistently specifies maximum cable lengths that are exactly half those allowed by the spec! And, worse yet, in the real world, if you have a cable configuration that theoretically should work, but the terminations are slightly bollixed, or there are slight impedance mismatches where cables join at connectors, the result is subtle and sometimes intermittent problems, rather than obvious failures.

Yeah, one big issue with multi-drop cables is for devices in the middle (or anywhere not at the ends) of the cable. They would be much more susceptible to reflections on the received signals, and also have the burden of driving two cable segments and not just one, so they would produce about half the signal strength on the incident wave that a device on the end of the cable would.

Vendors try to solve these problems by building in automagic self-configurating features, which help when they work and make things worse when they don't. For a while we had an issue with Adaptec cards, when they first introduced a feature they called "domain validation." This meant that instead of trusting what the device said its capabilities were, the card would dynamically test for certain device characteristics and capabilities. Our device would sit there and say "I'm synchronous, I'm synchronous, please please run synchronous so we can get the data transfer rates we need" and for some reason the card would say "Nyaah, nyaah, I don't BELEEEEVE you, say what you like, I'm going to run asynchronous anyway" making our stuff basically not work unless the end-user turned off domain validation. This is no longer true, but we never found out exactly what had changed about their "domain validation" process. Dpbsmith (talk) 19:21, 22 Jun 2005 (UTC)

Termination

The description of the termination network is inside out and backwards. Passive termination uses 2 resisters to TermPwr and ground. Active termination uses 1 resistor to a regulated 3.3 volt source. Differential is 2 lines that are each terminated. In single ended, the other line is a ground. The active termination uses less power and therefore uses a smaller resistance that more closely matches the cable impedance. The tutorial that is already link to has it right <http://www.scsita.org/aboutscsi/SCSI_Termination_Tutorial.html>. It should also be mentioned that the term power fuses have mostly been replaced by a current limiting device. ~ Dan Oetting 136.177.111.33 00:32, 5 November 2005 (UTC)[reply]

Thanks. I got that wrong so I have now corrected it. neilm, 5 November 2005.

What's used now?

It's been a long time since I used SCSI. What is used now in servers? Isn't some form of IDE used now? --Gbleem 16:13, 8 November 2005 (UTC)[reply]

Newbie Question

In a read or write command block is the data transfer length in bytes or LBAs?

JimT

Always LBAs. See the SCSI Read Commands article for more info.

neilm

Connectors

I miss the description of various SCSI connectors.

xerces8 --213.253.102.145 12:31, 16 January 2006 (UTC)[reply]

I have added a new page called SCSI connector

neilm 19 January 2006

Clockspeed ("SCSI interface overview" table)

2 Gbit isn't a clockspeed. In fact it's not a measure of speed at all. It's an amount of data. Clockspeed is measured in Hertz, and ONLY Hertz.

Agreed. The interfaces starting with SSA really do not fit in with the given column headings. Clock speed is mostly only relevant for the various iterations of SCSI Parallel Interface (don't quote me on that - I am really only familiar with SPI and iSCSI). Perhaps we should generalize the table and provide a decent overview of all of the SCSI transport protocols? Cy jvb 19:05, 22 February 2006 (UTC)[reply]

MB/s over Mb/s

I think it should be made more clear what speeds the different SCSIs run at. Most transfer rates are defined as bits per second (bps) where as what is described here (Bps) which i have come to associate as Bytes per second. This does not, how ever translate over well when comparing transfer speeds.

So either the MB/s needs to be changed over to Mb/s or a subtle hint needs to be added to the text. Is this reasonable?

yet another Matt 14:30, 22 March 2006 (UTC)[reply]

The word "Bandwidth" was not a good choice as a header for the column in the table. It is related more to clocking rates rather than the amount of data moved in a period of time. The same goes for "baud rate" and "transfer rate". The SPI-4 standards define "transfer rate" as the negotiated number of megatransfer per second. Megatransfers/sec is equivalent to MB/s only on an 8 bit bus. This is why Fast-20 Wide has a maximum throughput twice as much as Fast-20 Narrow. The usage of "transfer rate" in the specifications definitely does not match the usage in the table column. I have changed the table header to "Max. Throughput" to correct the context of the column.
As for the MB/s v. Mb/s issue - from my experience with dealing with storage devices, using orders of 8 bit bytes for measurements fits the storage context better than using orders of bits. Current SCSITA terminology for SPI devices (ex. "Ultra160") match the maximum interface throughput in Megabytes/sec. Since SPI was at first the one an only SCSI transport interface, it seems natural to continue the convention. Cy jvb 18:53, 22 March 2006 (UTC)[reply]

Disk size limits

What are the device size limits in SCSI ? This subject is always popular with ATA :-) Isn't there a limit at 2 TB (2^32 x 512 byte sectors) ?

xerces8 --213.253.102.145 11:00, 24 April 2006 (UTC)[reply]