| Title: | DEC Rdb against the World |
| Moderator: | HERON::GODFRIND |
| Created: | Fri Jun 12 1987 |
| Last Modified: | Thu Feb 23 1995 |
| Last Successful Update: | Fri Jun 06 1997 |
| Number of topics: | 1348 |
| Total number of notes: | 5438 |
I have tried talking to local distributor of Adabas but they
did not give me a straight answer as to whether Adabas,today supports
FULL AUTOMATIC CLUSTER RECOVERY as understood by us in RDB.
Could someone update me in this aspect of Adabas. Basically the
customer need to know how does ADABAS takes care of cluster failure.
And specifically how LONG does it takes to recover under various
circumstances such as nucleus node failure verses slave node failure.
There will be 100 active Adabas users.( you may consider 30 percent
on update transactions and the workload is about 2 tps)
By the way, does anybody knows the CPU overhead of ADABAS.
For example it takes 1 to 1.5 vup to transact one Rdb transaction.
I appreciate any response to this.
Thanks in advance, Jeffrey Choy
| T.R | Title | User | Personal Name | Date | Lines |
|---|---|---|---|---|---|
| 660.1 | Nothing | CLYPPR::BOOTH | What am I?...An Oracle? | Tue May 29 1990 16:11 | 6 |
As far as I know, Adabas has no automatic cluster recovery. Their
answer has generally been to the effect "Adabas can ceratinly do that
if you are willing to write the code." I don't think there is anything
in Adabas to facilitate cluster recovery.
---- Michael Booth
| |||||
| 660.2 | Some bits and pieces | NOT003::DENTI | Ian Dent, Nottingham, UK | Wed May 30 1990 12:06 | 24 |
Adabas sees a cluster as a collection of nodes connected via DECnet
and therefore treats them as a standard network. There is no cluster
recovery.
What Software AG recommend is that a generic batch queue with a
job limit of 1 is set up for the cluster and the nucleus is submitted
to this queue. A second nucleus is also submitted which waits in
the queue. Once the node running the nucleus crashes the second
job in the queue starts, accesses the database on the shared disks
and recovers any unfinished transactions. System then continues
running.
This is a simple explanation of something that would be hard to
get correct as the Net-work product must also be modified during
the startup of the "second" nucleus as this contains a list of which
node runs which nucleus.
Software AG plan to bring out a product called "Watchdog" which
will automatically detect a node in a cluster crashing and invoke
a command procedure specified by the operator.
To summarise, Software AG could talk around this subject enough
to confuse the customer but basically do not have any automatic
cluster recovery.
| |||||