Database Reference
In-Depth Information
capacity. However, I would have allocated 10TB in total, which I would use
for my distributed tables.
Remember, the distributed size is used only by tables that have been
specified as distributed. These are likely to be fact tables; so don't be
surprised to see a very large number here, a very significant percentage
of the total allocated. Furthermore, there is no replication of data in a
distributed table. This sets distributed tables apart from replicated tables.
In reality, both types of tables rely on disk RAID configurations for data
protection.
Log_Size
The log size value behaves similarly to the distributed size, in as much as it
is spread evenly across the appliance. Therefore, if I have 6 compute nodes
and a 10GB log file, I will split my log file allocation into 6 approximately
1.6GB allocations. Remember that SQL Server doesn't create a filegroup for
the log files.
In actual fact, the log file is split further again. PDW doesn't create just
one log file for the database. It creates one log file per disk volume instead.
With the HP AppSystem for PDW, for example, there are a total of 16
volumes per compute node. Therefore, a 10GB log_size specification for
our 6 compute nodes would actually result in 96 log files, with each one
being approximately 106.6MB in size. By doing this kind of allocation, PDW
ensures that the log file grows evenly across all its available storage and isn't
constrained to a single volume. Furthermore, each volume soaks up some of
the load from the database log. Typically, the log will be double the size of
the largest (uncompressed) file loaded.
Autogrow
PDW will run an autogrow on the database files rather than fail a query that
asks for additional capacity when the files are full. In PDW, the autogrow
option affects only the databases created on the compute nodes. This is
important because PDW's metadata via sys.database_files is a bit
misleading. This catalog view shows the information for the shell database
heldonthecontrolnode,nottheactualvaluessetagainstthecomputenode.
Table 10.2
and
Table 10.3
show the actual values set against the compute
node.
