Recently the MPG server was movd to a Tier 1 data center on a dedicated 100 megabit link. Ditto for WindInMyFace.com and the diglloyd.com image server.
Testing from another computer on a different link showed near theoretical max speed. As well, local testing showed that the server was delivering as fast as the link could go. Hence the server speed and link from the server out was/is really 100 megabit.
But at my home office which uses a Comcast business-class internet 100 X 20 megabit link, testing by downloading large files showed speeds typically around
20-40 megabits, far lower than the 100 megabits that the link ought to deliver, and that my server (see above) delivers from the data center.
At times speeds might briefly hit 60 megabits, but only very briefly.
See “Hops” towards the end of this piece for the explanation.
End-to-end and peering
When testing an internet connection for speed, most users go to a site like speedtest.net or similar.
These sites maintain fast peered links to most providers, so the speeds almost always look terrific to the customer: “I’m getting what I’m paying for”.
Testing internet speed almost always shows expected (fast) speeds because the test sites are wired-up (so to speak) to be fast against the networks of large ISPs.
It’s like testing how fast your car accelerates on a 20% downhill grade.
Comcast and other ISPs (Internet Service Providers) maintain peering* between their network and some very large providers (think Apple, NetFlix, etc).
That’s very worthwhile and good for accessing big popular sites, but for the vast majority of sites out there (like this one), real speed is less, even if the site is hosted on a high speed link.
* Peering as used here means semi direct connect links between a web site and an internet service provider in order to minimize latency and maximize bandwidth.
The WiFi hop
The foregoing assumed a wired connection from the computer to a fast cable modem or similar. But for
many users the use of WiFi is another “hop”, but a hop subject to speed degradation and latency issues ranging from modest to severe, depending on distance, interference and so on.
Thus for many users, WiFi is almost certainly the slowest and highest latency hop of all and thus the gating factor on internet speed (for large transfers).
In general, WiFi deliver very poor performance for small transfers in particular, such as file sharing of many small files. Wire computers into a gigabit switch instead which will benefit a local network tremendously as well as internet access.
Is the fastest internet speed worth it?
A fast internet link is only as fast as its slowest link.
Put another way, the quoted speed that Comcast or another ISP markets and hypes and sells is peak speed
unlikely to be realized for the vast majority of web sites.
On the flip side, that high speed really does deliver in some cases (for example, big software downloads from Apple and similar).
For most users a speed of 30 megabits or so is ample.
A 20/30/50 megabit link won’t be as fast a 100 megabit link when downloading, say, an Apple software update, but for most web sites the real world speed, the user experience, will hardly change. And after all, many web sites themselves are sluggish, so it’s not about connection speed at all.
A valid question when buying internet service is whether peak speeds for that occassional big download from a big popular site is worth paying for.
Real world speed depends on the investment in infrastructure
to deliver that speed consistently to web sites large and small. Then ask yourself if this QOS (quality of service) is likely to be a business goal of a large corporation serving millions of customers.
And then one understands the outcry over “paid peering” and the like (so-called “net neutrality”
efforts offer up serious negative long term issues, and are likely to be implemented as bludgeons, and must be handled very conservatively, if at all).
I investigated the behavior for this site and its sister sites and what I found is that my Comcast link has a slew of “hops” between my office and the server in the data center, meaning traffic is routed between many intermediate nodes before it can get to and from my web browser.
looks something like this using traceroute (some lines slightly truncated here, to fit one per line).
Note that the first eight nodes are Comcast (#1 is my cable model); my connection has to make 8 hops before it can even head towards my server.
UPDATE: I’m told by one reader that the hops are not the issue per se assuming no congestion (and no artificial gating or bandwidth controls or traffic prioritization, which one cannot rule out for Comcast). Whatever the reason, it’s clear that there are latency and bottlenecks involved, so the “on the ground” reality is that a 100 megabit connection is only theoretical for many web sites.
traceroute to diglloyd.com (18.104.22.168), 64 hops max, 52 byte packets
1 192.168.1.129 (192.168.1.129) 0.378 ms 0.296 ms 0.176 ms
2 * * *
3 c-73-170-32-1.hsd1.ca.comcast.net (22.214.171.124) 10.844 ms 8.725 ms 8.731
4 te-0-3-0-8-sur04.santaclara.ca.sfba.comcast.net (126.96.36.199) 9.432 ms
5 te-1-1-0-13-ar01.oakland.ca.sfba.comcast.net (188.8.131.52) 12.926 ms
te-1-1-0-9-ar01.oakland.ca.sfba.comcast.net (184.108.40.206) 10.918 ms
te-1-1-0-8-ar01.oakland.ca.sfba.comcast.net (220.127.116.11) 11.270 ms
6 be-90-ar01.sfsutro.ca.sfba.comcast.net (18.104.22.168) 13.664 ms
te-1-1-0-7-ar01.sfsutro.ca.sfba.comcast.net (22.214.171.124) 10.552 ms
be-90-ar01.sfsutro.ca.sfba.comcast.net (126.96.36.199) 16.406 ms
7 188.8.131.52 (184.108.40.206) 13.209 ms 15.673 ms 23.881 ms
8 be-17-pe02.11greatoaks.ca.ibone.comcast.net (220.127.116.11) 16.672 ms 16.510
9 snj-edge-03.inet.qwest.net (18.104.22.168) 16.390 ms 14.190 ms 14.535 ms
10 svl2-cntr-01.inet.qwest.net (22.214.171.124) 17.391 ms 17.059 ms 15.346 ms
11 126.96.36.199 (188.8.131.52) 15.006 ms 17.515 ms 16.497 ms
12 184.108.40.206.static.etheric.net (220.127.116.11) 21.869 ms 18.939 ms
13 * * *
14 18.104.22.168.static.etheric.net (22.214.171.124) 22.818 ms 22.409 ms 20.640 ms
With that many hops, it is not realistic to expect peak speed to my web browser from my server (via Comcast at least).