Need help configuring tape library for future Serviceguard installation

[Paul J. Ledbetter, II]

have two L2000s that are separately connected via HVDS SCSI to each drive of t
he 2/20 Ultrium Tape Library. I will soon be implementing clutering via Service
guard, installing OV Dataprotector on each node. I want whichever may be the ac
tive node to be able to access the robotic picker via Dataprotector. I know I h
ave to connect a cable to the library controller. However, I only have one libr
ary controller. How would I be able to configure my system so that no matter wh
ich server is active, Dataprotector would still be able to access the robotic pi
cker? My first thought would be to simply add a second library controller and t
hen extend the HVDS daisychain to each of the library controllers, one per serve
r. What do you think? How do other Serviceguard implementations work with a co
mmon tape library?

[TwoProc]

You've got to have a Cell Manager for Data Protector. Whichever machine is the cell manager would have the connection to the robot picker. If you've only got a direct connection (non-SAN), then I think you can only have one cell manager.

To make that connection float around, it would have to be in a SAN environment, with the SCSI connection for the picker connection to a fiber bridge, which is connected to the SAN - which either of the two machines could then connect to. However, don't let two machines connect (try to manipulate) to the picker at the same time. Of course, properly constructed, a service guard package would suit this nicely - but you're missing hardware layers.

[Steven E. Protter]

Shalom,

1 Controller, means one box can run the tape drive, not two.

I suppose you can put in a fiber network and make it available to both machines, but you'll probably have issues booting Ignite under taht setup.

For performance reasons you want to walk away from the daizy chain idea. These tape drives perform best when they are the only device on the SCSI chain.

SEP

[Paul J. Ledbetter, II]

John,
Would the servers be accessing the picker at the same time in an active/passive cluster configuration?

[Paul J. Ledbetter, II]

Remember, I am already connected to each tape drive individually. I am only trying to get HPUX to access the robotic picker when it is the active node. Why wouldn't a separate connection per server to its own library controller work? Wouldn't the active node only run the picker through its unique connection?

[Steven E. Protter]

Why?

The OS doesn't have the necessary components within to share the tape drive.

I've seen some filenames in ServiceGuard that might contradict that though.

Shared tape links and SG:
http://docs.hp.com/en/B3935-90015/ch01s02.html

This document looks most promising. It talks about shared tape services and mentions robotic arms.

You'll need to read it in order to evaulate its value to you.

SEP

[TwoProc]

Paul, You're OK with moving the cluster around, with only the active node bossing the robot around.

However, for your idea of using a second SCSI chain... it's feasible, but I don't know if it would fly. Upon reflection, it could be that what you're asking for is exactly what's done when a SAN is not present.

The best guy to answer this question (IMHO) is Scott McIntosh (pharoah). He probably doesn't review this forum very much, but he does participate heavily in the OV Omniback/DataProtector forum. He is world-class in this very topic.

So, you're next move I think is to repost to that forum, and hopefully he's not on vacation or something where he can't answer.

Or, maybe an admin will move this over to the correct forum before too long...

[A. Clay Stephenson]

Actually, you can even connect the tape drive to two hosts on the same HVD SCSI bus and you can also connect the robotics the same way. Obviously the drive (or robitics) can only be active from one host at any one time.

All you have to do is something like this:

Host A HBA SCSI ID 7 (termination on) ------ Tape Drive SCSI ID 1 ---- Host B HBA SCSI ID 6 (termination on).

Because the i/o load the robits add is absolutely miniscule you can daisychain the robot and a tape drive on a common bus (and adding inline terminators per MC/SG goodness rules):

Host A HBA SCSI ID 7 (termination off) ----- Inline Terminator ---- Tape Drive SCSI ID 1 ---- Robotics SCSI ID 0 ----- Inline Terminator ---- Host B HBA SCSI ID 6 (termination off).

You add the inline terminators (which look like short SCSI extension cables) for two reasons: 1) Allows replacement of an HBA while the bus remains in use 2) Being cautious, you simply don't trust the on-board termination of the HBA to be reliable when the node that houses it is powered off or when the HBA itself is defective.

By the way, this is not a theoretical discussion, I actually have some of my libraries configured like this and they have worked well for years.

[Paul J. Ledbetter, II]

Remember guys, I am not attempting to share anything between the servers. Each active node is expected to have sole control of the robotic picker through it own connection to a separate library controller. Now if there is something else that happens that happens behind the scenes on the inactive node then maybe I'm just screwed. That of course begs the question: how does everyone else do it without a SAN? The servers actually are connected to a FC-connected VA7400. That's how I plan to implement serviceguard. I just don't have the bridge or use any switches.

[TwoProc]

See A. Clay's posting above. It sounds to me like he does it exactly like you're asking.

[Paul J. Ledbetter, II]

I am not interested in connecting both hosts to a single tape drive as I have each host already connected to their own dedicated drive. However this scheme might work to connect two hosts to a single Library Controller. What I don't understand is why I wouldn't be able to simply connect a second Library Controller to an available slot in the back of the tape library and connect the second host to that second Controller. Then the active node in turn would have its own dedicated controller in addition to its own dedicated tape drive.

[A. Clay Stephenson]

Your scheme to use a 2nd library controller should work. I was simply going to the limit to show you that it is really not even necessary to dedicate a SCSI bus solely for a robot not is it necessary to dedicate a SCSI bus for a single drive accessible from only one host.

The advantage of connecting each drive to both hosts is that if a node fails then all of your tape drives are still available to DP (and if you've carefully configured your I/O paths so that for example /dev/rmt/c3t0d0BESTnb on hostA is also /dev/rmt/c3t0d0BESTnb on hostB (and likewise for the robotics device nodes) then NOTHING has to change when the DP package fails over to the alternate node.

You can do it your way but the the probability of the backup specifications running without alteration is less. With my approach, it's not necessary to test every time a backup specification is changed because the devices remain exactly the same.

I already been there and done that so I know it works and works well.

[Paul J. Ledbetter, II]

I'll try that solution and I'll tell you how it went. Thanks all for the input.