Vodafone Cable Speed Comparison: To Peer or Not to Peer

Vodafone Cable Speed Comparison: To Peer or Not to Peer

TrueNet noticed a few months ago that file download tests for TelstraClear Cable had wide variation in speeds during a single download, and frequently reached just 5Mb/s.

We have identified issues including the potential double impact of both not peering, and packet loss on a route between TelstraClear and our server. TrueNet's server is openly connected to both the Wellington and Auckland Internet Exchanges (WIX & APE) but these options were not used by TelstraClear.

Peering Exchanges

Peering exchanges around the world provide neutral interconnection that allows improvements in network performance between ISPs and websites peered at each exchange. The Internet in New Zealand has two main peering exchanges for ISPs; at the Wellington Internet Exchange(WIX) and the Auckland Peering Exchange (APE ).

Peering at these exchanges is common, for example by Trademe, banks, website hosting companies and news websites. The links above provide a list of peered ISPs and websites.

The decision not to peer is optional for an ISP. Vodafone peers under the name iHug, but TelstraClear did not peer at all, and at time of writing, their cable services remain not directly connected to WIX or APE.

Speeds over Cable

Recently, TrueNet studied TelstraClear Cable's performance for file download speeds on TelstraClear's 15Mb/s and 100Mb/s cable services. For a 300K file download, speed ramps up and then slows down. When testing a 1MB file download the maximum speed reached could be as low as just 5Mb/s, depending on the time of day.

The Internet without Peering

APE servers appear to be reached via an international route, while WIX servers are reached via any route available.

Without peering, an ISP directs traffic via whatever path they decide to use. Traffic might be routed “out-of-town”, or onto less "direct" networks with bottlenecks or other issues, which may slow file downloads to significantly less than advertised speeds.

For tests on TelstraClear volunteer probes, TrueNet's server is directly connected to WIX and then Catalyst, which connects to TelstraClear via FXNetworks, and a wholesale operator before finally reaching TelstraClear prior to heading to the volunteer's probe. This route apparently has severe packet loss.

TrueNet tests of TelstraClear Cable show these effects. The packet loss causes a TCP window size reset, which leads to the file transfer going through a “ramp-up” process again. This may occur multiple times during a single download, and is more evident when downloading larger files. Smaller downloads may have packet loss, but not enough to cause the window size reset.

Chart 1: Comparison of 300KB and 1MB file download speeds - June 2013

Click for big version.

Chart 1 shows that the 1MB file download, with speeds as low as 5Mb/s at 8pm, was consistently slower than the 300KB file download on both 15Mb/s and 100Mb/s services.

New Data

TrueNet has installed a new server at APE and is now testing cable probes from WIX, APE, Sydney and Dallas.

Chart 2: 100Mb/s Service. Download Speed Range and Average for Various File Sizes

Click for big version.

Chart 2 & 3 Key

For these tests, TrueNet measures the speed for each quarter of the file downloaded and reports the fastest.

Summary of Results

Multiple issues are evident in the results:

1. The National file is too small for packet loss to cause a reset, but is still affected by a slow ramp-up. It never reaches the maximum download speed of 100Mb/s and downloads at around 15Mb/s.
2. Speedtest CH: TelstraClear provide direct backhaul services to their customer Snap, who host the Speedtest server, enabling high speeds for that test.
3. Speedtest AK: downloads from CityLink are routed overseas and back so the results are almost the same as for Dallas and Sydney.
4. Akamai downloads have two distinct results; some probes get ~15Mb/s, others ~5Mb/s. This suggests two separate Akamai servers with different routes to each. This shows on Charts 2 and 3 as a large range.
5. Our 1MB Wellington tests for the 15Mb/s plan appear to get the speeds expected, but not during the busy hour as packet loss takes effect. As in Chart 1, both services are slowed to 5Mb/s.
6. Our 1MB Wellington tests for the 100Mb/s service run about the same speed as the 15Mb/s service. This is due to packet loss as described above.

Chart 3: 15Mb/s Service. Download Speed Range and Average for Various File Sizes

Click for big version.

The large range for the Dallas result is due to one Wellington probe that seems to have an excellent connection to Dallas and Sydney, otherwise the range is significantly less.

Since Vodafone took control of TelstraClear, TrueNet has assisted Vodafone to find these issues, and the specific causes of speed reductions. This included trials of various window sizes for TrueNet tests.

While lack of peering was not the cause of packet loss in these examples, it does make that type of issue more likely to be encountered, as non-peered traffic can traverse much less "direct" networks.

ENDS

© Scoop Media