LaCie 131018 - StorView Snapshot - Mac, Руководство пользователя

Страница: 194 / 214

StorView RAID Module
186
Maximum performance will be achieved when all drives are performing multiple commands in parallel.
To take advantage of a Full Stripe Write, the host has to send enough data to the controller. This can be
accomplished in two ways. First, if the host sends one command with sufficient data to fill a stripe, then
the controller can perform a Full Stripe Write. Alternatively, if the host sends multiple sequential
commands, smaller than a stripe size (typically matching the chunk size), the controller can internally
combine these commands to get the same effect. In the above example, if a 256K chunk size is used,
then the stripe size is 1MB (4 chunks * 256K). So, for maximum performance, the host could either send
5 * 1 MB write commands, or 20 * 256K write commands.
The effectiveness of the controller’s ability to perform a Full Stripe Write depends on a number of
parameters:
A.2 Sequential Access
If the commands sent from the host are not sequential, the controller will not be able to cluster them
together. So, unless each individual access is sufficient to fill a stripe, a Full Stripe Write will not occur.
A.2.1 Number of Outstanding Commands
For the controller to successfully cluster commands, there has to be a number of write commands sent
simultaneously. Setting the host to send up to 64 commands should prove adequate. Alternatively,
enabling writeback cache will have a similar effect, as the controller can then cluster sequential
commands even if the host only sends a small number of commands at a time.
A.3 Access Size
With very small accesses, it is necessary to have a large number of commands to cluster together to fill
up a full stripe. So, the larger the access size the better. It is best to use an access size that will fill a
chunk. Of course, even if a stripe is not filled up, small sequential writes will still benefit from command
clustering.
A.3.1 Access Alignment
The alignment of a command from a host system is determined by the command’s address. In an optimal
system, a write of one chunk of data would reside exactly within a chunk on one disk. However, if this is
not the case, this write will be split up into two separate writes to two different data drives. This will have
a negative effect on performance. To overcome these problems, the user can, with more sophisticated
operating systems, set the access size and alignment to an optimal value.
As can be seen from the figure below, to get the highest performance from this system, it is necessary to
have a number of stripes being written in parallel. As the array expands, with more and more drives, the
number of commands (and amount of sequential data) necessary to do this increases.