White Papers

28 PS Series Asynchronous Replication Best Practices and Sizing Guide | BP1012
7.3 Tuning the WAN link
When replicating across a WAN, the data packets will probably be traveling through a router. A router
generally has a memory buffer that stores incoming packets so that they can be processed and forwarded. If
the WAN link is congested (or too small) and it becomes a bottleneck, this can cause incoming packets to fill
up the memory buffer on the router. Eventually the router may be forced to discard (drop) incoming packets
until it frees up space in the memory buffer. This in turn causes the sending side to timeout because the
receiving side will never acknowledge receipt of the frames that were discarded.
If Group Manager or SAN Headquarters reports a high occurrence of retransmits during replication, this could
be due to an overloaded WAN link that is dropping packets. One course of action would be to monitor and
adjust the buffers in the router. Or, it may be necessary to implement Quality of Service (QoS) or Class of
Service (CoS) on the router or any upstream switches to reduce the amount of the data flowing into the
router. Adjusting router buffer settings is beyond the scope of this paper. Most manufacturers of these devices
include the ability to adjust parameters that can affect traffic flow. You should be aware of these capabilities
and use them if needed, particularly when replicating over slower speed link paths.
If the WAN link is not dedicated to the storage arrays, then QoS or CoS may be configured on a specific
VLAN, the IP addresses of the arrays, or even by port (3260, the default iSCSI port). Of course, any other
traffic that shares the link will also need to be managed and may affect the performance of the replication
traffic. If possible, you should use a dedicated link for the storage replication traffic.
Although it is also beyond the scope of this paper to discuss in detail, some customers have also utilized
WAN optimization products in their WAN links. These products may implement features such as compression,
deduplication, or other packet optimization that can help improve the efficiency of a slower WAN link, and
therefore decrease the time it takes to replicate changes across these links.
7.4 Planning for storage needs or volume sizes
The default, space-efficient guidelines for sizing replication reserves and delegated space are presented in
Table 4. The default values indicated are recommended for most situations unless the actual change rate is
well understood.
Replication space, default value, and space-efficient value
Replication space
Default value
Space-efficient value
Local reserve
(primary group)
No failback snapshot: 100 percent
Keep failback snapshot: 200 percent
5% + %Change_Rate
10% + %Change_Rate
Replica reserve
(secondary group)
200 percent (to ensure there is
adequate space for the last replica and
any replica in progress)
105%
+ %Change_Rate
x (# of Replicas – 1)
Delegated space
(secondary group
for all replicas
coming from a
single group)
Must be large enough to hold the sum
of all replica reserve sizes for all
volumes replicating to that group
Monitor change rate, adjust to lower
than default value, and continue
monitoring