Specifications

Parallel Data Warehouse Architecture CHAPTER 6 111
Control rack Data rack
Management node
active/passive
User queries
Control node
active/passive
Landing Zone
Backup node
Control rack
Active server Dedicated storage
Passive server
Data loading
Data backup
Dual Fibre
Channel
Dual
InfiniBand
SQL
SQL
SQL
SQL
SQL
SQL
SQL
SQL
FIGURE 6-1 The multi-rack system
The Data Rack
All activity related to parallel query processing occurs in the data rack, which is a collection of
compute nodes. Each compute node consists of a server with dedicated storage, a SQL Server
instance, and additional Parallel Data Warehouse software that provides communication and
data transfer functions. Although the compute nodes run separate SQL Server instances in
parallel to manage each distributed appliance database, you query the database as if it were a
single database.
The number of compute nodes in a data rack varies among the vendors, although each
vendor follows a standard architecture specication. For example, each data rack includes a
spare server for high availability. If a compute node server fails or needs to be taken of-
ine for maintenance, the compute node server automatically fails over to the spare server.
The current connections to the appliance stay intact while the appliance recongures itself.
Just as with SQL Server failover, queries that were in progress before the failover need to be
restarted.