SCSI termination question

[Gurm]

Here's the deal. Don't mix devices.

The scanner is probably a crappy SCSI-1 device. Even nice SCSI scanners are rarely better than SCSI2. Your hard drives are SCSI3, probably LVD.

You need to have the drives on the 68-pin UW channel, and the scanner on the 50-pin channel. End of story. Get a new ribbon cable that runs from the 50-pin directly to the rear of the case in order to do this.

- Gurm

[VJ]

Euhm, both segments will probably still form a single bus, so devices will still be mixed... The Adaptec 7890c is a single channel adapter. Only option is to add a second SCSI adapter, which is then solely used for the scanner.


Jörg

[MultimediaMan]

Jorg is right...

My Duallie has an onboard Adaptec 7890 Controller which is a dual channel UW (68 Pin only) adapter (Basically an Adaptec 2940U2W). I run my HDDs and UW SCSI CD-ROM from those adapters. My SCSI Burner and Changer are run from a PCI Adaptec 2940U 50 Pin adapter.

Whenever you use an 8 bit device on a 16 bit bus the entire bus segment defaults down to 8 bit.

[xortam]

Gurm is right. The P2B-xS MBs use an AIC 7890 which has a flex bus (or whatever their term is) which handles mixed SCSI levels w/o performance degradation. You land up with a single channel but two segmented buses that require a total of four termination points. I use both buses with my P2B-S.

[Gurm]

Danke, Xortam.



- Gurm

[xortam]

Slight correction/clarification Gurm ... the LVD drives should be connected to the 68-pin Ultra2 connector versus the 68-pin Wide SCSI connector. Check your manual Patrick ... it explains these various connections.

[Patrick]

Ummm... Ummm...

Ok guys, all this feedback is great. It's also interesting to see that everyone doesn't agree! No wonder I've found this SCSI stuff confusing.

I'll have to look at the manual carefully tonight and try to absorb what I read. What surprises me is that this computer was originally configured by one of the high end computer/video stores in town and it appears from some of what I've read in this thread that it's been misconfigured from the start. However, I don't want to change it unless I'm sure what I'm doing.

If anyone else has an opinion, please post it. This is quite educational!

[xortam]

Patrick ... You can verify your work once you've got your connections fixed. Go into the SCSI BIOS and turn on the verbose mode and you'll see the negotiated SCSI width and speed of each of your devices during the MB SCSI POST. Make sure those metrics match the manufacturer specs for each of your devices. Bus termination can be a little confusing at first even without this more sophisticated segmented bus functionality. I'll try to summarize it simply:

A SCSI bus has two ends.
SCSI buses must have both ends terminated and only the ends terminated. Anything else may sort of work but is unreliable.
The host controller can be anywhere on that bus: at one of those ends or somewhere in between.
Termination can be applied by jumpers, resistor packs, termination plugs, or at the host controller via BIOS settings, depending on the device's features.
Ultra2 buses require active terminators versus passive terminators.

Your MB uses two segmented buses so both buses must be terminated at each end when using multiple connectors.

[VJ]

Multimediaman: Thanks for stating that I'm right, but you say the controller is dual channel... I was stating it is single channel... perhaps some confusion here ?

The thing on a single channel is :

Single Ended devices can be mixed without loss of performance, LVD-devices connected to this bus will fall back to SE-mode. I was referring that if you mix different devices, you still need to be within scsi-spec concerning cablelength. So if your bus is 40 MB/s, and there are some 40 MB/s devices mixed with 10 MB/s devices, the 40 MB/s devices will run at full speed, but the cable is also limited to the maximum length allowed at this speed (I believe this then is 1.5 metres).
The number of segments that make the channel does not really matter.
On a 2940UW (not pro), only two connectors can be used at the same time. So you can have up to 2 segments on a single channel. Depending on the speed of the bus, you will have to make sure your cablelength is within limits. With the bus set to 40 MB/s, this is 1.5 metres; with the bus set to 20 MB/s, I believe it turns to 3 metres.

A dual-channel just behaves as 2 seperate controllers.


xortam1:
What do you mean with 4 termination points ? Could you make a small diagram ?

dt--d--d--host--d--dt

(d = disk, t=termination, host=hostadapter, this setup suggests using 2 connectors on the hostadapter, not positioning the hostadapter in the middle of a ribboncable or anything like that)
I only see 2 terminations in this setup (unless one half is 8 bit and the other half 16 bit, then the upper 8 bits will need to be terminated by the hostadapter).


xortam2:
You are right when you say that incorrect configs might work, but are unreliable. They can also work but at lesser performance. I do remember from discussions on comp.periphs.scsi that there is one exception: one single external narrow device connected to a host adapter needs not to be terminated. While this is a violation of scsi-spec, it is one that always works (small proof of this : most scsi-scanners don't come with a terminator, but with a host adapter; and it works). There is a theoretical explanations why it is ok, but I don't remember the full details.

Gurm:
You're right when you say devices shouldn't be mixed, but with proper interconnections (respecting the rule that every line must be properly terminated), it is possible to obtain a working setup, see the images on the Adaptec website I posted earlier. But I'll grant you that a lot of scanners do not behave well on a SCSI-bus. But all the additional converts cost quite a lot, and it is a drag to get things working, that it is simpler (and often cheaper) to buy a simple hostadapter for slow devices.


Jörg

(edited for spelling/layout)

[VJ]

Multimediaman:
-
Whenever you use an 8 bit device on a 16 bit bus the entire bus segment defaults down to 8 bit.
-

This is not true, I again refer to the link on the Adaptec page. It is possible to mix 8 bit and 16 bit devices on a single bus (even on a single segment !), you just have to make sure that the upper byte of the 16-bit devices are terminated where needed. This does result in the fact that the order in which the devices are placed can make a difference (to save on converters, etc.).

Jörg

[VJ]

xortam:
The manual of the IBM Ultrastar 36LZX (which is an LVD-drive), uses the Adaptec host adapter 2940UW as an example.
See:
http://www-3.ibm.com/storage/hdd/tec...56A780052CAC5/$file/ddys-dpss_digw.PDF

It therefore seems to me that Gurm is right when he says to connect the LVD-drive to the 68 pin UW connector (the 2940UW does not support U2-scsi !)


Jörg

[xortam]

VJ ... the P2B-DS supports up to Ultra2 Wide devices so the LVD drives should indeed use the Ultra2 connector (I'm assuming the ST118273Ws are LVD drives). Here's the P2B-DS manual so you can read about the segmented bus configuration, etc (pg. 33). Post again if that doesn't answer your questions. Patrick won't have to get involved with hi/lo byte terminations if he configures his SCSI the correct way as Gurm and I are recommending. Things can get a bit more complex if he adds some new SCSI variants into the mix. I don't want to cloud the issue too much or we'll overwhelm poor Patrick even more. Next thing you know we'll start discussing the issues surrounding Ultra640.

Here's a graphic showing the segmented bus ...


Other P2B-xS SCSI info.

[VJ]

xortam:
I should have read that manual... I thought it did not support LVD-drives (and hence not Ultra2Wide, as they just named the connector Ultra2). My mistake . LVD drives should indeed be connected to the Ultra2 68 pin.

But I think in this discussion, we're overlooking something : Patrick doesn't have LVD drives. The Seagate Barracuda model ST118273W is plain UltraWide :

http://www.seagate.com/support/disc/specs/scsi/st118273w.html

The LVD-model is the ST118273LW. Further evicence is the fact that Patrick states the drive has a termination jumper, which is never the case on a LVD-drive (also mentioned in the text on the above link).
Patricks drives should therefore be connected to the UW 68 pin...


(I think we're reaching some consesus here ! About SCSI !)

Jörg

[Patrick]

Can't stay up late tonight...

Jörg, you posted while I was typing this out. You are correct about the drives not being LVD.

I think I stayed up too late last night playing around with these drives. My brain feels like mush tonight.

Go into the SCSI BIOS and turn on the verbose mode and you'll see the negotiated SCSI width and speed of each of your devices during the MB SCSI post.

xortam, I thought at one time I saw the width of the drives listed during the SCSI BIOS post, but all I see now is the model numbers and a reference to 40 sync. I have verbose selected so I don't know why I no longer see more info.

Here's some other info if it makes anything clearer. Forgive me if I don't state all the technical terms correctly. These Seagate drives are single ended, have 68 pins, are 16 bit, 40 sync and are connected to the 68 pin wide SCSI connector on the P2B-DS motherboard. The scanner that is normally connected to the end of the same cable as the drives is an HP C6270A Scanjet.

I've downloaded a utility from the Seagate site called DiscWizard2000. It's supposed to help with jumper settings, etc.

I appreciate all the help, guys. I'm too beat to tackle this tonight but I promise to give a report when I have a clear head and can think properly. Thanks again everyone.

[VJ]

-
Forgive me if I don't state all the technical terms correctly.
-

No problem, later SCSI terminology tends to be confusing.

-
These Seagate drives are single ended, have 68 pins, are 16 bit, 40 sync and are connected to the 68 pin wide SCSI connector on the P2B-DS motherboard.
-

Good, that is where they should be.

-
The scanner that is normally connected to the end of the same cable as the drives is an HP C6270A Scanjet.
-

Is it possible to disconnect the internal cable from the component that goes to the outside of the case, and just have the internal ribbon cable left ?

If so, then you should try to put a harddisk on the last connector of this cable and terminating this disk. The other disk is then somewhere inbetween, not terminated. In this config, the disks should perform at their best (if the controller is configured correctly, which I suspect is the case, as it finds the drives and states their speeds).


Ideally, you'd want to connect the scanner to another controller, as connecting the scanner to this bus will most likely result in to long a cable. But that is a later step, first we'll try to maximize the disks' performance.


Jörg