Hi All

I have a client that's interested in a failover solution. They have
two servers on a corporate WAN connected via fibre. Although I have no
experience with RFS and/or replication it appears it could be an
effective solution.

Does anyone out there have any experience/war stories/gotchas re
implementing and using such things. If so i'd be interested in
anything and everything relevant I can get my hands on.

Rgds
Sue

Hi Sue

I have worked on a few DR projects and can give you some advice
on what not to do.

Any products which replicate block changes such as sndr or
rsync are fine , the database will replicate on another server
but the logical structure of the data will be lost whenever there's
a system crash . A good example would be a program that writes
master file records and updates indexes. The indexes might get
updated before the master records are written and then the server
crashes . The DR server will have the index entries on it but not the
data referred to by the indexes , the master records.
You need transaction logging of some form using logical transaction
boundaries. ie Start the transaction ,perform writes/deletes/amends
to the database in a logical group and then end the transaction.
Within that transaction there might well be 5 writes , 2 deletes and 3
amends
which must all happen together if the consistency of the database is to
be maintained.
If Unidata has transaction logging then the simplest form of DR is
to move the transaction log over to the DR server and update the
database with the log file. The DR side can be kept almost up to
date , depends on the time lag between the moves of the log files.

regards
Karl


rivensue (AT) hotmail (DOT) com wrote: