HP StorageWorks Scalable File Share System Installation and Upgrade Guide Version 2.2

Configuring SFS20 storage 5–29
5.4 Configuring SFS20 storage
You do not need to preconfigure the SFS20 arrays; all SFS20 storage is configured during system
installation, when you install the servers.
Proceed to Chapter 7 to install and configure the servers.
5.5 MDS LUN sizing guide
The size of the MDS LUNs used for a file system determines the number of files that the file system can hold.
If the default file distribution is weighted towards a large number of small files, the meta-data capacity will
dictate the ability to create files. As a result, even when the file system appears to have available space,
extra files may not be created. For this reason, HP recommends that the MDS LUNs are always sized
generously.
When sizing your MDS LUNs, use the rule of thumb of allowing 4KB per file; for example, a file system with
64 million files would require 256GB of meta-data storage. In addition to this rule of thumb, when planning
the size of the MDS LUNs, note the following points:
• Allow approximately 2GB for MDS overhead.
• As explained in Section 5.1.2, if a service LUN, another MDS LUN, or an OST LUN already exists in
a disk group on an EVA4000 array, an MDS LUN must be sized so that the total used size is not more
than 90% of the total capacity of the disk group.
This restriction does not apply to SFS20 arrays.
• Each OST LUN is configured such that it can have a maximum number of objects determined by the
following formula:
max objects = (LUN size/16K) objects
In some circumstances, this configuration can limit the number of files that can be created to less than
the limit imposed by the size of the MDS LUNs. The following two examples describe how the MDS
and OST LUN sizes interact in the case of large Lustre files and in the case of small Lustre files. In each
example, the system has 8 OST LUNs, each of size 512GB:
• Small files of 2KB are created.
In such a case, each file consumes one object on one OST LUN. As each OST LUN is of size
512GB, then according to the formula described earlier, each LUN can accommodate a
maximum of 32M objects and therefore 32M files. The eight OST LUNs can accommodate a
total of 256M files. The rule of thumb (4KB of meta-data per file) determines that the meta-data
capacity for this number of files is 1TB. The capacity of the OST LUNs limits the number of files
that can be created, even if the MDS LUNs are larger than 1TB.
• Large files, striped across all OST LUNs, are created.
In such a case, each file consumes one object on each OST LUN. Again, each OST LUN can
accommodate 32M objects. However, as each file uses an object on each OST LUN, the eight
OST LUNs can only accommodate a total of 32M files. The rule of thumb determines that the
meta-data capacity required for this number of files is 128GB. In this case, the capacity of the
OST LUNs is the limiting factor in terms of both LUN capacity and number of files, even if the
MDS LUNs are larger than 128GB.
If you expect that the predominant use of the file system will be to accommodate many small files, and
that the size of the MDS LUNs will be a greater limiting factor than the OST LUN capacity, increase
the size of the MDS LUNs.