Conference cookie::archive_backup

Mike,

Assuming you can handle the TAPESTART.COM correctly, this will a) work and b)
fly . I would make the node that controls the robot device (GKx200) the 
MDMS server node and ABS server node. If that node dies, you won't be able to
use MDMS/ABS and the TL896 anymore until the node comes up, but the other,
surviving, node can never reach the GKx200 device anyhow, as it's locally
connected to the died node. That's the disadvantage against connection behind
HSJ50's. The main advantage is performance (TZ89 drives can do 5 - 10 MB/sec...
and you won't get that with an HSJ50 unless to wanna blow away you CI-network)

Hans

Are you saying that even connected behind an HSJ50, if the wrong node fails
the TL896 will be unusable because of the logical connection?

Would it require a system reboot for the remaining server to take control of
the tape library?

If you connect a TL896 behind an HSJ50, that TL896 will be accessible by all
nodes on the CI. That's from the hardware point of view.
From the ABS point of view, you'll need to install the ABS-server on those nodes
in the cluster (= CI) you want to have as server. Mostly these will be the
departemental servers (4100's) and not the 8x00's (Enterprise) due to the costs
of licensing. 
In your case, you want to enable the 8400's as ABS servers, so you'll have to
buy 2 Enterprise server licenses, and install it on both servers; do a little
bit of post-configuration and you created a ABS-server, that runs on 2 nodes in 
the cluster. However, only 1 node will be the actual server and the other one
will take over if the first one dies. The same applies to MDMS and the Scheduler.

If you want to connect the TL896 behind HSJ50's you'll have to realize the
following :
1) a TL896 contains 6 (six) TZ89 drives
2) 1 TZ89-drive is capable of handling 5 - 10 MB/sec, which results in a 
   30-60 MB/sec flowing through your HSJ50.
3) One HSJ50 can only handle appr. 13 MB/sec if used with CIPCA's and 
   configured for performance.
4) One CI-network can only handle 13 MB/sec per path, so 26 MB/sec for both the
   A and B-path.
5) One CIPCA can also handle appr 15 MB/sec.

As you can see, there are a lot of performance bottlenecks. If you insist of
using HSJ50's, you'll need at least 2 CI-adapters, 2 Star-couplers and 4 HSJ50's
(2 redundant pairs) just for the backup. And the disks, containing the data,
must be on one or more CI-networks to get the data fast enough off the disks.
This is only valid for customers which no budget problem, as 6 KZPSA's are
about half of the price of one CIPCA.

All I want to say is "Yes, you can connect it behind an HSJ50 and No, you will
not require a system reboot, but don't expect anything called performance"

Hans

Mike & I are working the same opportunity as described previously in this 
note.

We have configured the TL896 with three TZ89s on each of two 8400s. There 
would actually be six KZPSA SCSI busses on each 8400 to handle a down system 
where all TZ89s would be connected to the 8400 that is still running. 
Obviously, there would some manual cable swapping in this case.

What we're trying to find out now is if its possible to have the robotics 
device (GKxxx) on a SCSI bus connected to the CI so that the robotics are 
shared, while the TZ89s are split as specified above for maximum performance.

I guess what we want to know is:

a. Will it work technically?

b. Has anyone ever tried it?

regards,

Jim Mickol
SBU Tech Support
Rochester, NY
716.436.2565

    I'll see if one of our robot utility people can answer this 
    question.  They are more familiar with the hardware setups.
    
    Chris

	You can't (easily) connect the robot to anything without
	also connecting one of the drives to that same place.  The
	recent TL820 family libraries have one SCSI connector for
	each drive.  There isn't an extra one for the robot.  Now,
	if you have two drives connected to one bus, that will free
	up a connector and you could have the robot on its own bus.

	So you could do:

		Bus ---- Drive - Robot
		Bus ---- Drive
		Bus ---- Drive
		Bus ---- Drive
		Bus ---- Drive
		Bus ---- Drive

	or maybe:

		Bus ---- Robot
		Bus ---- Drive
		Bus ---- Drive
		Bus ---- Drive
		Bus ---- Drive
		    ---- Drive
		   /
		Bus	
		   \
		    ---- Drive

	Mixing architecture types is something that has NOT been
	tested by ABS or any of the components it replies upon.
	Connecting the robot to an HSJ would allow equal access
	to robot by all nodes.  But our configuration testing has
	been limited to having the tapes on that same HSJ.  In the
	case of SCSI connected robots, the robot and all tapes
	have all been SCSI connected to the same host.

	We barely have the capability to test such a configuration,
	but I know of no plans to do so.  If you need something like
	this supported, contact our product managers; Judy Cross
	or Bryan Cox.

					Alan Rollow
					Media Robot Utility

I am looking at Figure 30 on page 4-38 of the TL82X/TL893/TL896 Facilities
Planning & Installation Guide.  It shows the MUC connected to TZ89 Drive 5.
Why couldn't the MUC be connected to its own SCSI and the six TZ89s be connected
to their own dedicated SCSI?  It looks like there are 3 unused connections on
the back of the TL896 that could be used for the MUC connection.

	It depends on how you have the tapes connected.  I don't have
	a six tape TL89x handy, but I'd guess it has six connectors;
	our 3 tape model has three.� It has been blindingly clear
	how the tapes are going to assigned to controllers.  If you
	want one tape per SCSI adapter (per bus), then all six
	busses are used.  If you go with two tapes per bus, then
	you have leftover connectors.  You will have to deal with
	the fact that the MUC is narrow vs. wide, but there's obviously
	somethere already.