This topic helps you determine the number of metadata servers that
are needed to handle the workload based on the application workload.
It is importance to understand the right parameters for loading
and sizing the SAN File System to help determine how many metadata servers
are needed for a given client workload. Important factors to consider are:
- Number and the mix of file system objects, such as files, directories
and symbolic links that would be involved in the combined workload as seen
by the Metadata server cluster, and how many filesets those file system objects
are partitioned into.
- Size and mix of the hotset of those objects and filesets that each client
would expect to operate upon with their respective applications.
- Typical file operations that a client application might generate. This
exercise requires expertise with the application under consideration. However,
FOPs for many well-known and standard workload classifications can be used
to estimate this information.
- An estimate for the type of caching that the metadata will have for the
given application load.
This estimate can be based on the mix of FOPs that the application
generates, the hotset of objects and the size of the metadata cache. However,
various client installable file systems might have varying amounts of memory
that can be used for caching the metadata.
Perform the following steps
to size the metadata servers based on application workload and characteristics
is based on the following steps:
- Understand the type of workload that the application generates. A reference
table for a variety of standard workload types has been presented to help
map your application to one or more of the standard application types.
- Gather specific workload and environment information by answering a set
of questions. Because it may be difficult to get answers to all the questions
for all the environments, we have divided the questionnaire into mandatory
and optional parts.
- Calculate the number of file operations that will be generated per second
for the specific workload and environment, based on the cache effectiveness
measure provided for various workload, and, based on that, calculate the number
of metadata server operations that will be generated.
- calculate the number of metadata servers required based on the number
of metadata server operations generated per second and the sustained rate
of transaction serving capacity that you assume for the metadata server.
The next few sections take you through each of the steps in detail.
Use
the Metadata servers — application workload method worksheet to help you size
the metadata servers.