"Tim Schaefer" <com> wrote in message
is noted in this PDF?
shared-nothing environment with DB2?
Not sure what you asking. A shared nothing environment simply means that
there is one partition per node (computer) so each partition has it's own
memory, disks, etc. So if you had 24 partitions, you would need 24 nodes.
These nodes are sometimes called "thin" nodes because each node usually
contains only one or two processors and less memory than most
multi-processor nodes. This works well with extremely large data warehouses
with frequent tablespace scans.
Often times, people will use clustering with multiple partitions per node
(and multiple CPU's per node). For example one might have 4 nodes, each with
8 processors, and each node with 6 partitions (24 partitions total). This
works better in transaction environments and data warehouses that are not
massive (and do not require frequent table space scans). In this
configuration, special care must be taken to have plenty of disk subsystem
throughput, and the placement of data on the disks is more critical than a
shared nothing partitioning system.
A shared nothing architecture scales in a more linear manner than a system
with multiple partitions (and multiple processors) per node. But this linear
scaling is more applicable to queries that do tablespace scans, rather than
a large number of smaller queries.