I'd be worried about a particular vulnerability in this scenario.
If the system goes down while you are updating overflow or oversized data
frames, you could get file corruption. A single U2 logical write can take
multiple os writes, and at any given moment on a busy system, you probably
will have this condition. You can minimize this risk by minimizing overflow,
but it's probably impossible to eliminate it.
Outside of uv replication, I don't think that there's a working solution to
this "in-flight transaction" problem. With Unidata, RFS handles it nicely.
-------
We use LifeKeeper from Steeleye Technologies to replicate linux
partitions at the OS level for UV redundancy. The servers are right
next to each other and replication is done over 3 gigabit NICs - 1 for
each of 2 partitions, and an extra for redundancy. There are crossover
cat5 cables running directly between the NICs. The users connect to a
virtual IP address that points to a 4th NIC on whichever node is
currently active. In the event of a failover, LifeKeeper handles moving
the virtual IP address, mounting the filesystems, and the startup of
applications. We have yet to experience a failover on the current
hardware, but I found during pre-installation testing that UV was
running on the backup box within a minute or so of failing over. YMMV.
LifeKeeper allows an application dependency hierarchy to be created so
you can define the order in which applications start. In our case, UV
depends on access to Active Directory (samba/winbind) as well as having
the underlying filesystem mounted.
LifeKeeper is built on top of linux software RAID, so /proc/mdstat on
the active node has the following contents:
Personalities : [raid1]
md1 : active raid1 nbd9[1](W) sdb2[0]
24065370 blocks super non-persistent [2/2] [UU]
bitmap: 2/184 pages [8KB], 64KB chunk, file:
/opt/LifeKeeper/bitmap_ntdata
md0 : active raid1 nbd1[1](W) sdb1[0]
80381196 blocks super non-persistent [2/2] [UU]
bitmap: 28/614 pages [112KB], 64KB chunk, file:
/opt/LifeKeeper/bitmap_uvdata
while the backup node shows no mounted RAID filesystems. If the primary
were to crash, the backup server would mount its local copy of the
filesystems and the primary server's replicated filesystems would be
marked as dirty. The primary would have its unmounted filesystems
synced to the backup when it came back online.
As others have pointed out, if you have a cold-spare scenario like this,
there is no add'l licensing charge from IBM. You do have to obtain a
2nd license in order to install UV on the 2nd server, but you won't be
charged.
-John
-------
u2-users mailing list
u2-users@listserver.u2ug.org
To unsubscribe please visit http://listserver.u2ug.org/
-------
u2-users mailing list
u2-users@listserver.u2ug.org
To unsubscribe please visit http://listserver.u2ug.org/
