White Papers

16 PS Series Asynchronous Replication Best Practices and Sizing Guide | BP1012
5 Test methodology
Because actual WAN links can vary greatly, the Netropy 10G WAN emulator was used to simulate the WAN
connection. The Netropy 10G uses dedicated packet processors to ensure precision and repeatability.
Besides throttling bandwidth, it can also inject latency, jitter, and packet loss. This allowed simulating the
behavior of several different kinds of SAN and WAN links between sites. The Netropy 10G was configured to
participate in flow control and to set a queue size (analogous to a router’s internal buffer) of 1024 KB (or 1
MB).
Performance was initially measured using a 10 Gb connection path with no WAN emulation. Once the
baseline performance across a 10 Gb link was measured, the WAN emulator simulated speeds of 1 Gb, OC3
(155 Mbps), T3 (43.232 Mbps), and T1 (1.544 Mbps) networks.
WAN speeds
WAN connection
Speed
10 Gb Ethernet
10 Gbps
1 Gb Ethernet
1 Gbps
OC3
155 Mbps
T3
43.232 Mbps
T1
1.544 Mbps
Actual WAN links can vary in speed or guaranteed bandwidth from provider to provider. Each test case used
one of the values shown in Table 2 with the WAN emulator to simulate the bandwidth of a WAN or SAN
connection between replication partners. Next, each step of the bandwidth throttling exercise was combined
with a random packet loss of 0.1 percent or 1.0 percent and latency (each direction) of 10, 20, and 50ms. The
results of each combination were recorded for comparison to the previous runs in which there was no
additional packet loss or latency.
Before each replication, I/O was run to the volume to simulate changed data. In some cases, 100% of the
data was changed, and in other cases, only a portion of the volume. A replica was manually created each
time. When three volumes were replicated simultaneously, a replication group was created allowing
replication to start on all three volumes at the same time. This study measured the effects of the following
parameters on asynchronous replication:
 RAID level for the primary and secondary groups
 Single volumes compared to multiple volumes
 Thin provisioning
 Connection bandwidth
 Packet loss
 Latency and the TCP window
 Pool configuration