Detect a new tape device, OVMS 7.2-1

[John T. Farmer]

Hi all,

I'm adding an external DAT tape (with auto-loader) device to our Alpha server. Can someone point me to documentation or discussions about getting that new unit reconized by VMS. The unit is in addition to, not replacing the existing internal DAT tape device. I was reading about SYSMAN which appears to deal with system device configuration. Help is appreciated.

Thanks,

John

[Steven Schweda · Accepted solution]

SYSMAN> help io auto

IO

AUTOCONFIGURE

This command is for use on Alpha systems only. It automatically
identifies and configures all hardware devices attached to a
system by connecting devices and loading their drivers. On VAX
systems, use the SYSGEN command AUTOCONFIGURE.
[...]

[Hoff]

Hot-adding or hot-removing a SCSI device isn't necessarily goodness -- you can corrupt SCSI bus traffic doing this. (A few SCSI devices do have a "bus quiesce" button, but most don't.)

I usually prefer to shut down to add or remove SCSI devices. (This on non-testing and production servers.)

A default and typical OpenVMS bootstrap will find and autoconfigure the device, assuming no address collisions or hardware problems.

And yes, SYSMAN IO AUTO/LOG is your ticket.

[John T. Farmer]

I will add that this is the only device on the SCSI controller, so I was less concerned about doing this HOT. The test box already has it's device installed but I will try testing the SYSMAN utility there before attempting on the production box where the new device is installed. Not at a point now, that we can reboot the machine due to some long running jobs.

Thanks,

John

[Steven Schweda]

> I usually prefer to shut down to add or
> remove SCSI devices.

Sissy. I suppose you turn off the power
before [dis]connecting cables, too. And wear
a grounding wrist-band. Where's the
excitement?

[John T. Farmer]

Can I be confident that SYSMAN IO AUTO will not modify or disrupt any currently installed devices, and only pickup the new one? It appears to have no affect on devices on my test system, but just wanted another confirmation before I try on prod.

Thanks,

John

[Jan van den Ende]

John,

>>>
Can I be confident that SYSMAN IO AUTO will not modify or disrupt any currently installed devices, and only pickup the new one?
<<<

Weel, if it gives you some peace of mind: we have had to use it quite often in the mare than 10 years our cluster is now running, and never noticed any ill effects, not on v6.2, 7,1, 7.1-2, 7.2-1, 7.3-1, nor 7.3-2

I would guess that is enough for you :-)

Success.

Proost.

Have one on me.

jpe

[Hoff]

I've seen SYSMAN IO CONFIG mess up a few times, but it's very rare. I've certainly seen it miss hot-added devices. Messing about with the SCSI (parallel) bus itself is orders of magnitude more likely to cause badness in my experience, either with transients on the bus or the occasional crash or system lock-up, or with a failure to configure. Some devices really want to see a SCSI reset for best flavor -- the driver may well choose to toss a bus reset, though.

I should add: if the device is configured and functioning properly. If the device isn't configured properly or isn't working right, then all bets are off. (There was one SCSI driver bug a while back where a certain user-level operation with a standard SCSI device could wedge all of OpenVMS. And that didn't involve breaking the SCSI chain, nor using SYSMAN. But I digress.)

I've fried equipment or data when I haven't followed precautions, and I've recovered from same when others haven't.

But to answer your SYSMAN question: Can you get away with this? Probably. Would I stake a production run on a critical server on it? (Would I recommend it or support it?) No. But if you've already broken the SCSI chain to hot-add a device, well, go for it.

SAS (and SATA) reduces various hassles here: bus addressing, termination, and bus transients. Some complexity returns if you're using SAS fanout, but I digress.

[John T. Farmer]

I must show my hardware ignorance by asking, what is "breaking a SCSI chain?".

I used an empty SCSI port on my DS20E Alpha. I terminated the SCSI chain on the back of the Compaq DAT Autoloader tape unit, then plugged the SCSI cable in between the tape unit and Alpha SCSI port and powered on the tape unit.

I had hoped Sysman could be run to recognize the tape device. I didn't really find a parameter to have Sysman act on a specific SCSI port so as not to modify anything else that is already functioning. Is there an approach to allow that?

I have a series of jobs running that can't be interrupted, so I would have to delay a reboot of the system by a week.

Thanks,

John F

[Hoff]

SCSI is a daisy chain bus. Daisy chains can be subject to transients when you insert or remove a device; reflections, transient gaps.

Some gear can selectively quiesce the bus temporarily, basically holding up the traffic while the bus is reconfigured.

You can usually get away with breaking the SCSI chain, since you usually don't hit a packet and most devices don't trigger reflections.

In a production environment, I'd tend to not recommend hot-add or hot-remove, unless you have a storage controller with a quiesce option. (And that's usually only external storage controllers; I've not seen a PCI controller with a quiesce.)

The device autoconfiguration process within OpenVMS occurs during bootstrap (unless it was explicitly disabled), and the configure process (and the USB management tools, on V8.3) are started and these will detect and offer up remote served devices (or any USB widgets) that arrive after bootstrap. Local devices that arrive after the bootstrap -- other than selected cases -- are not detected without manual intervention. (eg: SYSMAN IO AUTOCONFIGURE)

Some storage I/O hardware available can indicate hot or warm add is permitted, but the host software may or may not deal with that in an entirely appropriate fashion. The better grade gear tends not to throw bus transients, which is goodness. But once in a while, you end up with a mis-set or incompatible or busted widget, and you can basically crow-bar the whole bus, or (less often) sometimes the host.

In my experience, I prefer to avoid making hardware changes to production gear that's currently involved in production. (I might well plan for how to better allow for such changes, such as implementing clustering or application fail-over or an external SCSI storage controller (with RAID, quiesce, etc) other such (re)configuration or run-time changes, but that's fodder for another discussion and another thread.)

Stephen Hoffman
HoffmanLabs LLC

[John T. Farmer]

Ok, I what I'm hearing is there is not a way to direct the SYSMAN util to scan and detect a device for a single SCSI port (where only one device is connected), as opposed to scanning and configuring the whole SCSI bus. So I will wait for a time to do a shutdown and reboot to cleanly add this device.

Thanks,

John

[Steven Schweda]

> Ok, I what I'm hearing is [...]

Did someone say that? SYSMAN IO AUTO has
/SELECT and /EXCLUDE options, too. I don't
know if that makes it fool-proof, but it
might help. SYSMAN HELP IO AUTO /SELECT

> I used an empty SCSI port [...]

If there's only one device on that SCSI bus,
and you limit the scan to that bus (which is
pretty much doomed to be idle, as the OS
knows of no devices on it), then it looks
pretty safe to me.

Some real expert might know for sure that
/SELECT does what I'd expect (or not). I
suppose that it _could_ scan everything and
then ignore most of the results, but I'd
expect it to be smarter than that.

[John T. Farmer]

Well, the bad news is I had an unexpected system shutdown today (HP is still reviewing the crash dump), but the good news is after the reboot, I see the new tape device. So, now I am glad I took the cautious advise of some, and did not run the sysman on my busy live system.

Thanks to all,

John

[Anton van Ruitenbeek]

John,

Lucky for you, but there are some situations you can't add the device on a running system. This has to do with some devices shuts the port during power-on if there aren't any cables (or turminators) connected to it. This can be done for not used ports.
Any other cases - if there aren't any other devices connected to it, and your sure the cables are the right one (SE/DIFF/LV etc) its easy to do in if the system is up and running. For some reasons also read 'Hoffs' notifications and be aware of ESD. I did it for the last 15 years always 'hot' on production machines and till now it always worked. :-))

AvR