September 16, 2013
Prospects for Hot-Plugging HP 3000 Disks
I've had many A-Class and N-Class systems. I've always used them with fiber-attached disk. I am wondering about the internal disk drives. Are they hot-pluggable?
My objective here is to find a better alternative to DLT and DDS tapes for offsite storage. I've had suggestions of DS2100 and Jamaica drives. But a few 300GB Ultra SCSI drives would hold a lot more data with less points of failure. I intend to set up a BACKUP_VOLUME_SET and use the internal disks to do store-to-disk backups of the system.
Jim Hawkins, formerly the IO maven for HP 3000 systems at HP, replied with details.
There are multiple layers of changes for actual hot plugs or swaps to work.
- You need the disk HDD to handle this electrically.
- You need HDD physical carrier and physical interface to comply.
- You need the system physical interface and receptacle to comply.
- You need your Host System Bus Adapter (HBA) to electrically support this.
- You need the OS to be aware enough of the HBA to not get flustered by absence of the device and deal with any notifications from the HBA of the activity.
Given that the N-Class disk cage has a screw-based cover and the HDD carriers have no quick release levers (as compared with HASS/Jamaica or VA7400) I would state definitively that there is no hot-plug intention. At the same time, the SCSI bus is pretty low power and low voltage, so it would be generally not too unsafe to experiment. But you're also close to AC inputs and they are not low power.
Hawkins took the time to answer the question with some theoretical possibilities.
Might you be able to pull/push a drive where you've closed the volume? Likely it would work, but there may be all kinds of noise and stress on the SCSI bus which may not be well handled. However, I think each disk is on its own HBA channel which isn't shared with anything else, and so unlikely to abort someone else's IO.
This takes us to the last issue: mechanical wear.
These connectors were likely intended for more or less permanent mating of two components. Very likely they have a limited number of cycles that they are specified to hold-up. I've seen connectors that are specified for fewer than 25 cycles before you lose gold contact material. This is okay for normal HDD where one might replace one or two per slot in a system lifetime, but not sufficient if you're doing nightly back-ups and swaps. Connectors, where there is an expectation of a high number of pull/replace cycles, have special designs.
Now a little good news here is that the N-Class was still pretty much old-school HP design, so likely they didn't pick up something cheap that saved them .2 cents per unit on gold plating. No idea though if the HDD connector is a 10-, 100-, 1000-cycle part. Your system, your risk.
Mark Ranft, who posed the pluggable question, pointed out that the HP design choices for 3000s seemed to make the servers and their components good candidates for exceptional wear.
It is especially helpful to understand the concept of mechanical wear on the connectors. HP always had excellent and innovative hardware engineering on their HP 3000 (and HP-UX) servers. Remember, you can drop them off a building and still self-test them.
I've been doing some digging and I found the following link to the HP-UX forum. The Unix N-Class appears to allow hot-pluggable drives.
The actual power supply and the fans are in the front of the N-Class. The power receptacles in the back have internal cords that lead to the front.
No more trying to figure out what runs on
MPE/iX or where to find it. No more worrying
about availability! www.MPE-OpenSource.org
is all things MPE/iX: Open Source packages,
freeware, scripting, plus loads of tools
and information to keep your 3000 system
alive and thriving!
The comments to this entry are closed.