Gateway E-842R, Руководство пользователя

Страница: 56 / 164

CHAPTER 4: Configuring your storage solution www.gateway.com
50
Configuring the array writeback cache
In a writeback cache operation, data is sent to the controller from the host.
Before actually sending the data to the drives, the controller immediately
responds to the host confirming the data was received and written to the disk
(even though the data may have not been written to the disk). The host may
then send more data. This can significantly increase performance for host
systems that only send a low number of commands at a time. The controller
caches the data. If more sequential data is sent from the host, the controller can
cluster the writes together to increase performance further. If sufficient data is
sent to fill a stripe in RAID 5/50 configurations, the controller can perform a Full
Stripe Write, which significantly reduces the write overhead associated with
RAID 5/50.
Disabling writeback cache makes sure that the data is sent to the drives before
status is returned to the host. With writeback cache enabled during a short term
power failure, the battery back-up unit provides adequate power to make sure
that cache is written to disk when the power is restored.
In duplex operations, the cache is mirrored to both controllers, which provides
further redundancy in the event of a single controller failure. Mirrored cache is
designed for absolute data integrity. The cache in each controller contains both
primary cached data for the disk groups it owns and a copy of the primary data
of the other controllers. Mirrored cache makes sure that two copies of cache
exist on both controllers before confirming to the operating system that the
write operation has completed.
Normally, write-intensive operations benefit from the higher performance
when writeback cache is enabled on that array. Read-intensive operations, such
as a streaming server, may not benefit from writeback cache.
Initializing the array
Initializing an array clears all the data from the drives. This ensures the validity
of the data and parity stored on the array drives.
Two features of initialization are background and parallel . After the array is
created, initialization automatically begins in the background. While
initialization is in progress, you can create logical drives and the disks are made
immediately available to the operating system where data can be loaded.
Important
A trusted array does not calculate parity
across all drives, so there is no known
state on the drives. As data is received
from the host, parity is calculated as
normal, but it occurs on a block basis.
There is no way to guarantee that parity
has been calculated across the entire
stripe. The parity data is inconsistent, so
a drive failure within a trusted array
causes data loss. Before you use a trusted
array in a live environment, you must
initialize it.
As arrays are created and the initialization begins, you can have up to 64 arrays
initialized in parallel at the same time. You can also stop the initialization, or
pause then resume it at a later time. The controls for managing are displayed
on the
Main window next to the Array_Name after the initialization has started.
If you stop an initialization, the array is automatically Trusted.
You can initialize the array at a later time in which you could choose the option
to trust. This option should only be used in environments if you fully understand
the consequences of the function. The trust option is provided to give
immediate access to an array for testing application purposes only.