HP Hitachi Dynamic Link Manager Software User Manual

Page 43

2. HDLM Functions

was distributed will be used. However, if a specified number of I/Os has been

issued to a path, processing switches to the next path.
If the I/O is not sequential with the previous I/O, these algorithms select the path

to be used each time an I/O request is issued.

• Extended Round Robin

The paths are simply selected in order from among all the connected paths.

• Extended Least I/Os

The path that has the least number of I/Os being processed is selected from

among all the connected paths.

• Extended Least Blocks

The path that has the least number of I/O blocks being processed is selected

from among all the connected paths.

Table 2-4: Features of the Load Balancing Algorithms describes the features of the

load balancing algorithms.

Table 2-4: Features of the Load Balancing Algorithms

#
Some I/O operations managed by HDLM can be distributed across all, available paths,

and some cannot. Thus, you should be aware that even if you specify the Round Robin

algorithm, some of the I/O operations will never be issued uniformly across all the

given paths.
The default algorithm is the Extended Least I/Os algorithm, which is set when HDLM

is first installed. When an upgrade installation of HDLM is performed, the algorithm

that is currently being used is inherited.
Select the load balancing algorithm most suitable for the data access patterns of your

system environment. However, if there are no recognizable data access patterns, we

recommend using the default algorithm, the Extended Least I/Os algorithm.
You can specify the load balancing function by the

dlnkmgr

command's

set

Algorithm features

• Round Robin

• Least I/Os
• Least Blocks

These types of algorithms are most effective when a lot of discontinuous,

non-sequential I/Os are issued.

• Extended Round Robin
• Extended Least I/Os
• Extended Least Blocks

If the I/O data is from something like a read request and is generally sequential

with the previous I/Os, an improvement in reading speed can be expected due

to the storage system cache functionality. These types of algorithms are most
effective when a lot of continuous, sequential I/Os are issued.