An investigation of data throughput between offices.
A 10Mb file was created:
dd if=/dev/zero of=datatest bs=10000000 count=1
This file was transferred between the source host and the destination host using FTP. There are three interesting hosts on the network: zipcar, zipval, zipshack. zipcar and zipshack are both business ADSL accounts. zipval is a residential ADSL.
The first test compared a direct file transfer to a transfer through a PPTP tunnel:
zipshack -> zipcar, direct = 74kB/s, 76kB/s
zipshack-> zipcar, PPTP = 74kB/s, 73kB/s
zipcar -> zipshack, PPTP = 66kB/s, 67kB/s
The next test compared a direct file transfer to a transfer through an IPSEC tunnel:
zipval -> zipcar, IPSEC = 40kB/s
zipval -> zipcar, direct = 41kB/s, 43kB/s
zipcar -> zipval, direct = 44kB/s
Recognizing that throughtput tests vary based on multiple factors, the results indicate that tunneling (PPTP, IPSEC) has very little impact on throughput, and that the zipval host has half the throughput of the others.  Comparing ADSL bills for the three locations shows zipshack and zipcar are using what others have called ADSL 3.0, while zipval is using ADSL 1.0. Not only was the zipval service slower, it was more expensive, offered less bandwidth and was costing more in data overages every month. Time to upgrade!
The service provider promised no service interruption during the upgrade, but still managed to bring down the connection for 36 hours while they did “testing”. They sent a new wireless gateway (Siemens Gigaset 567) that “had to be installed before upgrading the service”. The new gateway came with software that does not run on Windows XPSP3 or Vista, and in fact is not supposed to be bundled at all with this gateway – “it’s plug and play” – which in fact is is not: it was impossible to register the MAC of the new gateway without calling the provider. After about 4 hours on the phone with technical support, the new gateway was online.
So the question: replace the Sonicwall router with the new gateway, use the new gateway with the Sonicwall as a DMZ, or toss the new gateway and use the old DSL router. More throughput tests are required:
zipval -> zipcar, direct, old modem: 87kB/s
zipcar -> zipval, direct, old modem: 90kB/s, 94kB/s
zipval ->zipcar, direct, new modem: 93kB/s, 99kB/s
zipcar -> zipval, direct, new modem: 74kB/s, 71kB/s
The results above were obtained with the new modem and a laptop as the only devices on the network. The old modem was tested through the Sonicwall router with other devices on the network, which I think explains the slightly slower speeds. The DMZ scenario, despite similar speeds, was discounted as it introduced between 2ms and 5ms additional latency. Since a primary goal of improving throughput is to handle VOIP traffic, it made no sense to introduce an additional hop, not to mention all the NAT issues.
A handy tool for checking VOIP is TestYourVOIP.com. Prior to the upgrade, a test from zipval to Montreal scored 3.8. After upgrading, the score was 4.1 (out of 5).
