|
|
Archives of the TeradataForum
Message Posted: Mon, 19 May 2003 @ 14:06:01 GMT
Subj: | | Re: Buddy AMP/Buddy Node |
|
From: | | El Naggar, Sameh |
A buddy AMP was a term used in the Teradata V1 environment i.e. DBC/1012 and 3600 Systems where we used the Teradata Operating System
(TOS). An IFP which is an Interface Processor (PE today) used to boot TOS from a Buddy AMP's disk. Also an IFP used to dump to the Buddy
AMP's disk when it crashes. The first AMP that started used to be the buddy of the first IFP that started and the second AMP buddies with
the second IFP and so on. IFPs needed buddy AMPs because they didn't have disk storage assigned to them for booting TOS and dumping. We
don't have buddy AMPs today on Teradata V2 systems.
Buddy nodes, today, is a term used in Teradata V2 for UNIX MP-RAS only. It is the mechanism used to protect from node failures and allow
vproc migration within a clique. Here, a Buddy Node is that which backs up FSG segment modifications of another node in the same clique.
For example, in a 4 node Clique with Nodes A, B, C and D, node B is called the buddy backup of node A, meaning that any modified segments on
node A will be shipped to node B across the bynet. Similarly Node C will be the Buddy backup for Node B, Node D will be buddy backup for C
and finally A will buddy backup D. If one node crashes, for example A crashes, or panics, then we don't loose the modified segments on A
because during the restart, Node B (its buddy) will flush the segments belonging to A to disk. The buddy node has to be on the same clique
so it can access the disks of the crashed node, to be able to flush. Buddying is automatically done when Teradata starts or restarts. In the
previous case if Node A stays down, Teradata will restart and the buddying will be done now with only three nodes. The vprocs that were
supposed to start on node A will now migrate and be distributed on the other 3 nodes in the clique.
Thanks and regards,
Sameh
| |