FTTH is a coming to our homes. Let's see what to expect in real-life, and the caveats to look for.
In this article we'll see :
. a Gigabit Ethernet performance guide : understanding and using Gigabit Ethernet
. the testing and evaluating of the FTTH Internet access of an European ISP which offers 1GB/s download, 200 MB/s upload
Last edited December 31 2015
1. Model network
For this study, I'll use a simple home network, with two inexpensive SOHO equipments : an additional switch and a router to connect two other rooms :
Switch 1 is a TP-link TP SG-105E ( 30 € )
Router 1 is a D-link DIR 818-LW ( 60 € )
2. Gigabit Ethernet
Gigabit Ethernet offers 1Gb/s ( ie 1 billion bit/s ) at layer2 ( ie including preamble and complete Ethernet frame ).
The copper wiring standard used in home networks is 1000BASE-T ( sometimes incorrectly referred to as 1000BASE-TX )
a. Gigabit Ethernet requirements
To achieve Gigabit Ethernet speed, we need :
. all communicating nodes to use Gigabit Ethernet
. each network segment 100 meters (330 feet) long max, using at least CAT 5 cables.
b. Gigabit Ethernet throughput
Gigabit Ethernet being 1 billion bits/s at layer 2, let's see the real usable payload.
From a 1538 bytes Ethernet frame, we can subtract :
- Ethernet : Preamble, header, FCS and inter-frame gap 38 bytes
- IPv4 header 20 bytes
- TCP header 20 bytes
We can too compute how many packets per second ( pps ) a gigabit ethernet link should at least allow :
pps = 109/8/1538 = 81 274 pps
Here is the resulting bandwidths, in an easy chart :
It is important to consider these values, when using network speedtests and benchmarks
c. Gigabit switches real throughput
a gigabit switch performance is described by several values. Here are the specs of a simple 5 ports switch, the TP-link TL-SG-105E :
Packet Forwarding Rate 7.4Mpps
MAC Address Table 8K
Packet Buffer Memory 2Mb
Jumbo Frame 16KB
One of the most important metric is the Mpps value. Although it is a marketing value ( doubled because of full-duplex, computed using a best-case scenario : long 1538 bytes frames, no options that could raise the processing needs, .. ), it casts some light over a switch performance.
Here, with 7.4 Mpps, we can assume 3.7 Mpps half-duplex.
Considering the value of Gigabit Ethernet = 81 274 pps we just calculated above, we have :
3 700 000/81 274 = 45.5 gigabit streams
this 5 ports switch shouldn't be limited by its packet forwarding rate.
We'll benchmark our test switch in the benchmarking chapter below
d. Gigabit routers real throughput
A gigabit router performance is described by two main metrics :
. its switching side performance ( Lan side )
. its routing performance ( Lan -> Wan, Wan -> Lan )
not all SOHO products manufacturers will publish these values. In the case of the Dlink DIR 618 LW used for this test, we don't have them.
We can look at SmallNetBuilders to have some datas ( LAN to WAN Throughput, WAN to LAN Throughput, Total Simultaneous Throughput, Maximum Simultaneous Connections ) :
The point here is that a gigabit router won't automatically be able to route at gigabit speed.
We'll benchmark our test router in the benchmarking chapter below
e. Gigabit hosts real throughput
network end-devices ( computers, NAS, .. ) obviously need gigabit capabilities, but this won't be enough.
Here are some limiting factors to consider :
. processing power ( we can assume 1Hz per bit/s, so 1 GHz will be enough )
. bus limitation, bus sharing ( a network interface connected via PCI or USB2 will be limited )
. OS overheads ( OS firewall, antivirus software may slow down the effective bandwidth )
. storage media ( a NAS or a PC downloading test may be slowed down by its HDD speed )
f. Gigabit infrastructure cabling real throughput
One rule of thumb is to avoid running electric wires and Ethernet cables in parallel. Yet if they do, they should run 1cm away per metter of parallel run ( ie if they run in parallel for 2 meters, they should be at least 2 centimeters away from each other )
Infrastructure cabling should be checked for any defect ( bent or pinched cable, electro-magnetic interferences ) that could lower the effective performances.
Ethernet plugs wiring must follow several strict rules and guidelines. Clumsy home-made Ethernet cables can lower the cable performance. Use them only if you feel competent ant confident enough.
Beside visual inspection, network benchmarking is an effective way to verify an infrastructure cable can achieve its designated performance level.
3. Gigabit Ethernet benchmarking
We'll benchmark our test network, using iperf3.
Iperf3 measures the throughput of the TCP or UDP payload. As computed above, 1 Gbit/s ethernet = 949 Mbits/s of TCP payload. ( notes (1) (2) ). Thus, 949 Mbit/s will be our baseline.
I'll keep the best result out of 5 ( as long as the results are logical and coherent ).
How to do a simple network benchmark between two computers using Iperf3 :
We need to download the Iperf3 portable folder on the two computers at https://iperf.fr/iperf-download.php
One PC will act as an Iperf3 client, the second one as an Iperf3 server
1. On both computers, we un-zip it
2. we open a command line, CD-in the iperf3 folder, and run :
server : iperf3 -s # CTRL+C to quit
client : iperf3 -c 192.168.1.20 # we use the Ipfer3 server's IP here
Note 1 : Choosing which PC is the server is important :
The Iperf3 client needs to be able to reach the Iperf3 server through TCP 5201 ( routes, routers firewalls/NATs, Iperf3 server OS firewall )
Note2 : Instead of Cd-ing in the Iperf3 folder, we can use a full path :
TL SG-105E switching performance
Let's first benchmark the Switch 1 performance, by running iperf3 between PC3 and PC4 :
average : 946 Mbits/sec
peak : 950 Mbits/s
we're really reaching the theoretical limit ( 949.3 Mbits/s ). we couldn't expect the switch to perform any better.
DIR 818LW switching performance :
Let's benchmark the Router 1 switching performance ( ie LAN to LAN ), by running iperf3 between PC2 and PC1 :
average : 946 Mbits/s
peak : 949 Mbits/sec
Here too, we're reaching the theoretical limit ( 949.3 Mbits/s ). The switching performance is really OK.
DIR 818LW routing performance :
Let's benchmark the Router 1 routing performance, by running iperf3 between PC2 and PC4 :
average : 487 Mbits/s
peak : 501 Mbits/s
Here we clearly have a cap. The Dlink DIR 818LW won't route above 501 Mbits/s. This will clearly limit our network performances.
Cabling performance :
Let's finally test any infrastructure Ethernet cable, to be sure the cabling is healthy ( no pinched or bent Ethernet cable, no electro-magnetic interference,… ).
Here need to check the long Ethernet cable between Switch 1 and the Internet Gateway.
Unfortunately, this long Ethernet cable can't be unplugged conveniently enough from the CPE, so I can only test PC3 to PC5 bandwidth ( which mixes-up the cable performance with the Internet Gateway switching side performance ).
Let's run an Iperf3 between PC3 and PC5 :
average : 938 Mbits/s
peak : 948 Mbits/s
Clearly the cabling is good, and is not limiting our gigabyte ethernet performance.
I strongly encourage you to test the infrastructure cabling level of performance.
4. FTTH presentation
FTTH ( Fiber To The Home ) is the state of the art technology for high-bandwidth Internet.
Each client gets its own private fiber, unshared, from the optical Main Distribution Frame ( Optical Exchange Node ) to the client's wall-plug.
The technology used by the ISP tested here is 1000BASE-BX-10 ( 802.3ah ).
1000BASE-BX10 is capable of up to 10 km over a single strand of single-mode fiber, with a different wavelength going in each direction. The terminals on each side of the fibre are not equal, as the one transmitting downstream (from the center of the network to the outside) uses the 1,490 nm wavelength, and the one transmitting upstream uses the 1,310 nm wavelength.
IEEE 802.3ah, also named Ethernet in the First Mile (EFM), is a June 2004 IEEE 802.3 standard (Ethernet).
It defines three types of links :
. point to point using copper
. point to point using fiber
. point to multipoint using EPON fiber
in our case, we're using 1000BASE-BX10 :
“ 1000BASE-BX10 defined in clause 59, providing point-to-point 1000 Mbit/s Ethernet links over an individual single-mode fiber up to at least 10 km. “
“ 1. What is Ethernet in the First Mile ?
Ethernet in the First Mile (EFM) is the nickname of IEEE Std 802.3ah-2004, an amendment to the Ethernet standard, specifying “Media Access Control Parameters, Physical Layers, and Management Parameters for Subscriber Access Networks”. The EFM standard was approved by the IEEE Standards Board in June 2004, and officially published on 7 September 2004.
The “Last Mile” is the name traditionally given to the part of a public communication network that links the last provider-owned node (the central office, the street cabinet or pole) with the customer premises equipment (CPE). The “First Mile” is the exact same thing, viewed from the customer's perspective.
EFM does not improve or replace the existing Ethernet. It is a set of additional specifications, allowing users to run the Ethernet protocol over previously unsupported media, such as single pairs of telephone wiring and single strands of single-mode fiber (SMF). This makes the EFM port types suited for use in subscriber access networks, i.e. the networks that connect subscribers to their service provider. “
( source : http://www.ethernetinthefirstmile.com/faq_gen.html )
the benefits of FTTH are :
. high bandwidth
. low latency
. immunity from electromagnetic fields
FTTH patch cable
the patch cable used by this ISP to link the CPE to the optical wall plug is a 9/125 SC-APC/SC-UPC :
FTTH usage precautions :
the only requirement at home are :
. to avoid angling the fiber cable : we need a fiber cabling running straight or in wide curves.
. to keep the SC-APC/SC-UPC plugs clean, avoiding any dust or grease.
. to never try to look directly through a live fiber cable ( the infrared light won't bee seen, but it may cause eye injury )
5. FTTH benchmarking
Let's finally test our FTTH performances.
Of course we'll perform our tests right behind the Internet gateway, to avoid any additional performance loss.
So we'll be using PC 5 :
Here are the tests that were performed :
Iperf3 download # Iperf3 download test using a public Iperf server
ipv6-test.com/speedtest # IPv4 speedtest
speedtest / lyon la fibre # www.speedtest.net using Lyon / LaFibre server
speedtest / paris freemobile # www.speedtest.net using Paris / FreeMobile server
mire.sfr.fr # SFR speedtest
testdebit.info # testdebit.info speedtest
direct 5 GO # Browser 5 GB file download using IPv4 on testdebit.info
wget -O /dev/null … # wget -O /dev/null http://1.testdebit.info/fichiers/5000Mo.dat
( Linux only )
The OS tested were :
Windows 10 + Microsoft Edge
Ubuntu 14.04 live DVD + Mozilla Firefox
( For some tips about Network Benchmarking with Windows and Linux OS, see 7. Notes / Network Benchmarking OS tips )
The tests were repeated many times, looking for repeatability and coherency.
The goal was to test the highest reached performance, so I kept the best score, as long as it was consistent with the global pool results.
Here are the final results :
Ubuntu live DVD is really performing best. Of course, there was no Antivirus or firewall running on Ubuntu, so we can't really blame the Microsoft OS architecture.
The values above are the results of a downloading test, hiding the highs and lows during each test :
Under Ubuntu, iftop indicated some peaks at 984 Mbits/s during the tests, so we really have a gigabit fiber link !
The only oddity is the Iperf3 test, that kept capping at 200Mbits/s, whatever the settings and the tested Iperf public servers.
Testing FTTH is a really subtle task, as many factors come into account :
. OS tested, firewall, antivirus, CPU power, HDD
. ISP infrastructure sizing : is the ISP collect network and core network wide enough ?
. ISP peering links sizing : are the ISP peering links wide enough ?
. ISP speedtests cheats : is the ISP prioritizing the speedtest packets and websites ?
Overall, with a lot of efforts, we can only assert that we really have a 1Gbit/s link, and that we can use a significant enough portion of it during real tests.
6. Is Fiber FTTH really worth the run ?
There are several cases were Fiber Internet access really shines :
. Single user experience ( A little speedy browsing shows some 20+ Mbits/s peaks, saturating an ADSL2+ line. Add-in a little webTV player and we start experiencing slowdowns. )
. Several simultaneous users in the local network.
. Heavy downloads ( OS iso, torrent, … )
. Uploading ( while I didn't cover this specific issue, 200Mbits/s of upload speed really is an asset )
. Low latency ( web surfing really feels snappier )
For all of these reasons, Fiber Internet really is worth it.
As for FTTH fiber, as long as the local network has been optimized for it, it really pushes the limit higher, leaving a lot of additional room for simultaneous users/activities/downloadings, giving a really snappy and instantaneous experience !!
Network Benchmarking OS tips :
It is not easy to benchmark the Gigabit/FTTH throughput as the firewall or antivirus software will lower the performances.
Using Windows, we need to start in failsafe mode with network support, to have the firewall and antivirus disabled ( which leads to security issues that make me not use this solution )
Using a Linux OS, we can use a fresh live Ubuntu DVD ( which comes with no firewall by default )
we edit the repositories :
sudo gedit /etc/apt/sources.list
# we add the universe and multiverse repositories
sudo apt-get update
sudo apt-get install iftop # command line bandwidth monitor
sudo apt-get install flashplugin-installer # flash player, needed for some website speedtests
sudo iftop # launches the command line bandwidth monitor
one very interesting Linux option is the ability to save to /dev/null, removing any HDD cap :
wget -O /dev/null http://1.testdebit.info/fichiers/1000Mo.dat