-----Brian Laks wrote: -----

>I have 9 LTO-4 drives connected to our TSM server, only 5 of them
>work at any time after upgrading to a new fiber card.
>
>We have tried two different fiber cards (qlogic and emulex) and still
>only 5 drives.
>
>We have Confirmed latest drivers and firmware with IBM support.  The
>old 2gig fiber card still works with all 9 drives, but is
>considerably slower than 5 drives on an 4g card.
>
>The new cards are both multi port cards, but only one port is being
>used.  Multipath drivers are not being used.  9 drives are zoned to
>one port.
>
>IBM support believes it to be hardware since the old card works, so
>we purchased a second card of different manufacture.  Now the problem
>exists on two fiber cards of different manufacture so I'm real
>reluctant to think its a hardware problem any more.
>
>Interestingly, the 5 good drives vary.  I can unload the drivers and
>reload everything and drives that were previously unavailable work
>while drives that were working are then unavailable.  It seems kind
>of random.  All the dives show up in the OS, and TSMDLST show them
>all as well.  I uninstalled old drivers and reinstalled them exactly
>as per IBM support instructions, and rebuilt the drive and library
>paths in tsm during each reload as per IBM support guidelines.
>
>Has anyone seen anything like this?  I'm absolutely baffled.  I'm
>thinking we are going to have to zone 5 drives to one port and 4
>drives to the other, but the SAN admin type is reluctant since all 9
>drives work with the original card.  My guess is that somewhere in
>the drivers its smart enough to know that 9 LTO-4's to a single 4g
>port is silly in the first place.
>
>Maybe someone wants 4 LTO-4's so this problem just goes away :)  The
>5 drives work with fewer problems and better throughput than the 9
>drives on the old card.

We have seen similar behavior during our last two DR tests. We run
our TSM server under mainframe Linux. When we have a DR test we
recreate the server at a SunGard hot site, using a mainframe,
SAN, and tape library provided by SunGard. During each of the last
two tests, we were able to get every tape drive working at one
time or another, but never able to get all of them working at the
same time. Unfortunately, we do not have a tested fix for the
problem, or even a conclusive determination of the underlying cause.
Some of our staff suspect that incompatibilities between the SAN
configuration and the virtual machine definitions are a factor.

Our Linux system runs as a guest under zVM at the hot site.
Roughly speaking, zVM is the mainframe analog of VMWare (although
I usually think of VMWare as the Intel analog of zVM; zVM and its
predecessor products have a history going back decades further
than VMWare).