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

Enabling server read cache 9–5
9.2.4 Setting the lustre.groups_ssh_timeout attribute
Whenever there is a timeout on the ssh connection to a group server, an event such as the following is sent
to the event log:
group_upcall: Failed to get groups for uid 100, timeout waiting to connect to
172.100.100.100
You can search for such events using this example query:
sfs> show log facility=lustre && data contains “timeout waiting to connect to”
If you see this happening frequently, you can increase the timeout from the default 5 seconds as follows:
sfs>
set attribute lustre.groups_ssh_timeout=8
You must stop and restart all file systems for this change to take effect. The upcall mechanism caches group
data so that in most cases even if a timeout occurs, the operation will continue using the cached value.
9.3 Enabling server read cache
By default, when an OST service reads data from disk, the data is not cached. Enable a read cache on the
OST services only if both of the following conditions apply:
• The applications that are predominately run on your systems reread the same file or files from multiple
client nodes.
Note that read caching always occurs on the client node, so that if the file or files are only being read
and reread on the same client node, enabling a read cache on the servers provides no benefit.
• The aggregate size of the files involved is smaller than the aggregate memory available on the
Object Storage Servers. Note that not all of the physical memory on the Object Storage Servers is
available for caching the objects.
If the applications that are predominantly run on your systems read and reread large files, keep the
default setting of no read caching on the HP SFS system.
Two attributes are used to control the read cache operation and size:
• Setting the enable_readcache attribute to yes enables a read cache on the OST services.
To enable the read cache, enter the following command:
sfs> set attribute enable_readcache=yes
• Setting a value in the readcache_size attribute controls whether objects are cached or not.
The read cache is best used to fully cache small objects that are read frequently, and you can use the
readcache_size attribute to restrict the use of the read cache to small objects. The
readcache_size attribute (which is set on a per-OST basis) is set to 32MB by default, so that when
the read cache is enabled, objects of less than 32MB in size will be cached in memory at each OST
service. You can change the value of the readcache_size attribute, which is specified in bytes.
A file may be striped over one or more OST service. When the size of file content on any OST service
grows to more than the limit specified by the readcache_size attribute, that OST service will no
longer cache the file in memory. Given a typical stripe size of 4MB, any file of less than 4MB in size
will be completely stored on one OST service, within one object, and there will be a direct
relationship between the size of the read cache and the file size for those files.
To change the value of the readcache_size attribute, enter the following command. Specify the
value in bytes:
sfs> set attribute readcache_size=limit_value
When you make a change to the enable_readcache or readcache_size attributes, you must reboot
all Object Storage Servers to bring the change into effect.