This topic helps you determine the number of metadata servers that
are needed to handle the workload based on the number of hard disk drive available
to SAN File System.
This method for determining the number of metadata servers looks
at the number of physical data disk drives in the SAN File System configuration
and performs a calculation from this number. This method relies on the fact
that hard disk drives can provide a fixed amount of I/O per drive. The more
drives, the more I/O throughput that can be obtained. As the number of I/Os
is scaled up, the number of metadata server transactions is also scaled up,
which has a direct correlation to how many metadata servers are required.
The
calculation for this method is:
metadata servers = ( hard disk drives / 50 ) * application factor
where:
- Hard disk drives
- Number of physical hard disk drives for the data (all drives in all storage
pools except for the system pool). This includes physical drives but not logical
drives. If there is a RAID 5 LUN that consists of 5 drives, this should be
counted as 5 drives. Hot spares or standby drives should not be included in
this number. If you are using RAID 5 and a 5 + P array, you must count all
six drives because the data is striped across all six drives.
- Application factor
- A factor from 0.5 to 2.0. If the application is cache friendly for the
SAN File System client, use 0.5 as the application factor. When the application
factor is 0.5, we use one metadata server for each 100 hard disk drives. TPC-H,
data warehouse and similar applications are in this category. If you know
the application is client cache unfriendly, create-and-delete intensive, or
if you want to be conservative, you might use 2 as the application factor.
In this case, each metadata server supports about 25 hard disk drives. Otherwise,
use 1 for the application factor.
Use the Metadata servers—available storage
worksheet to help you size the metadata servers.