HP 12000 Design Manual page 114

Figure 51 Dividing the Backup Jobs by Priority Level
On the target device you also have the ability to limit replication traffic beyond limiting the
replication time windows. This can be done by controlling how many concurrent replication jobs
are allowed per replication target, and optionally controlling the maximum Mbytes/sec throughput
of a single replication job. (This is set as a global configuration value that applies across all
replication targets on that device.) In this way, the amount of replication traffic can be controlled
to each replication target on a target device.
Setting the maximum number of concurrent jobs per replication target needs to take into account
various factors such as VLS device hardware limits and the network latency of the replication link:
In order to get maximum replication throughput, you must scale the number of concurrent
replication jobs per target depending on the number of nodes used for replication in the target
device. For example, if you have a 4-node VLS configured as a single replication target, you
should set the maximum number of jobs for the replication target to be 28 (based on four
nodes each with a maximum of 7 concurrent replication jobs per node).
There is a hardware limit of seven concurrent replication jobs per VLS node, so if you have
multiple replication targets on a VLS (such as many-to-one deployment) you need to ensure
that you do not exceed the hardware limit when you configure the "Maximum Simultaneous
Transfers" on each replication target.
For example, if you have four replication targets configured on a 2-node VLS, you do not
◦
want to set each Target to allow seven concurrent jobs because this would add up to 28
concurrent jobs across all four targets but you only have 14 concurrent jobs capability
in the 2-node configuration. So if all the replication windows were the same across the
four source devices, you would potentially starve some of the replication targets. (If you
had non-overlapping replication windows then there would not be any contention issues.)
◦ Another example would be an active-active configuration where you had a 1-node VLS
device, because then you have to share the hardware limit of seven replication jobs across
both the outgoing and incoming replication jobs. In this case you could leave the default
setting of four concurrent jobs in the target thus leaving at least three jobs for the outgoing
replication traffic.
The network latency can affect how many concurrent replication jobs are needed in each
replication target. The VLS replication incorporates latency-acceleration technology, but with
longer network latencies this still requires some level of concurrent replication jobs to saturate
the replication link as shown in
1 14 Replication
Table 22 (page 115).