White Papers

21 PS Series Asynchronous Replication Best Practices and Sizing Guide | BP1012
6.6 Packet loss effects
Asynchronous replication is also affected by the quality of the link between replication partners. If a link is
dropping packets, the asynchronous replication processes will have to resend those segments. When packets
are unacknowledged (lost), TCP/IP protocol invokes an algorithm known as slow start (see RFC 5681, TCP
Congestion Control). Slow start is part of a normal congestion control strategy to avoid sending more data
than the network or other devices are capable of handling. However, if too many packets are dropped and
slow start is invoked too often, it will affect the throughput of the network and slow down replication.
Figure 9 demonstrates that the addition of a small amount of random packet loss (1/10th of a percentage)
caused only a slight variation in the time it took to replicate 10 GB of modified data. However, one percent of
random packet loss was added, replication time doubled. Because packet loss can have a significant effect
on replication time, it is important to monitor the quality of the link between sites.
Effects of packet loss
SAN Headquarters can be used to monitor retransmission rates and set alerts when high levels of retransmits
are detected. In a normal and healthy iSCSI SAN, the percentage of retransmits should remain below 0.5
percent. While there are other contributors to the cause of retransmits, a noisy WAN link or a misconfigured
device somewhere in the path could cause packet loss and lead to slow replication performance.
11.6
12.45
24.82
0 5 10 15 20 25 30
0
0.1
1
Replication time (minutes)
% packet loss
Effects of packet loss across an OC3 (155MB/s) link (replicating 10 GB)
10GB data