T.R | Title | User | Personal Name | Date | Lines |
---|
3619.1 | New FW. | SUBSYS::TRAN | Straight <Left> Hitter.. | Thu Jan 16 1997 18:46 | 4 |
3619.2 | | CRONIC::LEMONS | And we thank you for your support. | Fri Jan 17 1997 16:32 | 20 |
3619.3 | Multi-Unit-Single-Lun (MUSL) FW is 2A5A. | SUBSYS::TRAN | Straight <Left> Hitter.. | Fri Jan 17 1997 19:25 | 5 |
3619.4 | | NABETH::alan | Dr. File System's Home for Wayward Inodes. | Fri Jan 17 1997 21:20 | 2 |
3619.5 | | SANITY::LEMONS | And we thank you for your support. | Tue Mar 04 1997 16:51 | 11 |
| Thanks for the answers. I'm checking this with Bill Coots, but I'm
planning to have the MUSL code installed now, though I won't be using
two-unit support until at least the summer.
I read the instrutions on how to set up MUSL. But what are the
differences and advantages of Multi-Unit Single LUN (MUSL) over
Multi-Unit Multiple LUN? Is MUML the kind of multi-unit support
available in the near past?
Thanks!
tl
|
3619.6 | | NABETH::alan | Dr. File System's Home for Wayward Inodes. | Tue Mar 04 1997 18:46 | 18 |
| The advantage of MUSL is that it works and complies to the
SCSI-2 specification for medium changers, so that software
can support it. MUML presented each robot as a separate
logical unit of the same target. The individual element
spaces were distinct and had to be separately controlled.
By itself this was little different than have N towers
not bolted together. The problem was the pass-through
addressing wasn't SCSI compliant. To move a tape from one
side of the PTM to the other you had to use port addresses
that weren't contiguous, and thus something that no reasonable
SCSI-2 software implementation would expect.
Software using the serial interface wasn't constrained by
this, but that wasn't a very popular interface. I think
DECls used it.
The advantage of MUSL is that the robot handles making
all the element space appear as one large robot.
|
3619.7 | | SANITY::LEMONS | And we thank you for your support. | Tue Mar 04 1997 22:48 | 4 |
| An excellent summary, and exactly what I needed. Thanks for taking the
time to write it!
tl
|
3619.8 | to musl or !to musl | DECWET::RWALKER | Roger Walker - Media Changers | Tue Mar 04 1997 23:06 | 12 |
| For UNIX and NT the choice of one mutli-tower or several
single towers depends on your needs.
The multi-tower configuration has the primary
advantage that any tape can go to any drive. With only
three drive in each TL820/822 tower this can be significant.
Single towers allow simultanious operation. NetWorker locks
the complete unit while loading tapes, not just until the
tape is moved, but until it can verify that it's the right
tape by reading the first data record. So if you have a
lot of tape changes then single towers may be faster.
|
3619.9 | | SSDEVO::ROLLOW | Dr. File System's Home for Wayward Inodes. | Wed Mar 05 1997 15:24 | 10 |
| One general disadvantage to a MUSL configuration vs. single
towers in the current firmware, is that if one tower can fail
a Test Unit Ready command, the whole configuration fails the
command. This reduces the MTBF of the configuration. Or
at least this is how we were told it would work. I haven't
done anything to verify it.
For the VMS products, HSM is probably better served with
multiple independent towers. I'm not sure about SLS and
ABS.
|
3619.10 | | SANITY::LEMONS | And we thank you for your support. | Thu Apr 17 1997 19:16 | 24 |
| Hi
I'm planning a couple of things, that I'd like to sanity check here.
Near-term, I plan to add a second TL8nn to our NetWorker environment.
I'm planning to bolt the two TL8nn units together. Here are the goods
(+) and bads (-) I perceive:
+ any tape can go to any drive;
+ one virtual tape library, rather than two smaller ones;
- if one TL8nn fails, both fails [I may need more details on this]
others?
If I don't bolt the boxes together, is that a MUML configuration? Or
can MUML support bolted boxes as well?
Longer-term, I'm planning to have 4 TL893 boxes connected to a two-node
TruCluster running Digital UNIX V4.0c (which, I believe, introduces
sharing of tape devices) and NetWorker Vn.n (where 'n.n' introduces
support for this configuration). I'm very concerned about a problem on
one box hanging all boxes. Part of the work for this configuration
will be to back up SAP R/3 archived redo logs. I can't afford downtime
for all boxes at the same time.
Thanks in advances for your thoughts.
tl
|
3619.11 | | NABETH::alan | Dr. File System's Home for Wayward Inodes. | Thu Apr 17 1997 20:58 | 14 |
| MUML is the multi-tower configuration when the libraries
are bolted together, but each robot controller (MUC) shows
up a different logical unit of the same target ID. It had
the severe disadvantage of requiring moves between towers
to do Element -> PTM -> other PTM -> Element. This was
further complicated by the fact that the PTM addresses
don't follow the SCSI spec.
If you have separate disconnected towers, you have separate
disconnected towers. One slight advantage is that if you have
lots of robot operations going on, you can one such operation
going on at the same time for each robot. In a MUSL configuration
the requests are serialized even if they could be done in
parallel.
|
3619.12 | | DECWET::RWALKER | Roger Walker - Media Changers | Sat Apr 19 1997 01:14 | 13 |
| If you take care to make sure that tapes a distributed properly
between the units, several single units will be better.
If the tapes all end up bunched up you could watch one unit
stay busy while the others just wait. This isn't a problem
the the multi-tower mode.
If you do use single towers then you will need an IOD for each.
Considering how much time you spend managing your backups, the
cost in convenience of the multi-tower would not be worth
the decrease in robustness. This would be different for a user
that just plugs it in and leaves it alone.
|
3619.13 | | SANITY::LEMONS | And we thank you for your support. | Tue Apr 22 1997 17:46 | 19 |
| Hi
Thanks for the replies. I've also learned, from the NETWORKER notes
file, that NetWorker supports both MUSL and MUML, but (according to
Kevin Farlee [note 179.6]):
"If you run the MUML firmware, NetWorker will see it as multiple
independant jukeboxes, incapable of interchanging tapes."
So, even if the boxes are bolted together in a MUML configuration,
NetWorker won't be able to move tapes between the boxes.
So I'm pondering which is more valuable to me: one virtual jukebox,
where all media are accessible by all tape drives, or several
jukeboxes, each with a fairly dedicated and segmented pool of tapes,
but without a single point of failure (if one ATL fails, the others
keep on going).
tl
|