Sometimes it's working well, then it slows to a crawl.
I've had Verizon FiOS service since the first week it was available in our neighborhood (July 2005). At that time it was advertised as 15 Mbps down and 2 Mbps up, a.k.a. 15/2, for $49.99 per month. Measured speeds actually came very close, as shown here. And these measurements were repeatable.
Over the years the service has been upgraded by Verizon with no action required on my part. Today my service is called "FiOS Internet 20/5." Unfortunately the price has also crept up, to $62.99 per month, again with no action required on my part!
The problem is, today, my FiOS service is highly variable. At times it's very slow. Skype audio breaks up on some calls and works fine on others. But this is anecdotal. Where's the data?
Gathering data
So ten weeks ago, I started a regular program of testing using a keyboard macro program on a mostly idle PC connected via a wired Ethernet cable to a switch and then to the FiOS service jack. I set it up to run sequences of 3 Speedtests and 3 Pingtests every two hours throughout the day. In our house, these tests seldom ran uninterrupted for more than 30 hours and my manual restarts were sometimes delayed by several days. Also, the accuracy of the macro software wasn't always up to the fine mouse pointing that Speedtest and Pingtest require to select servers. But over the past 10 weeks I've managed to get 646 good Speedtest runs against nearby servers (in Boston) and 549 good PingTest runs against Boston servers.
This evening I've done some data analysis which confirms the wide variation I saw anecdotally.
Next I looked at hour-by-hour averages to see if the performance variation depends upon the time of day, e.g. perhaps everything slows down in the evening... But no, that's not it.
This is the average of all measurements in each one hour period (24 hour clock).
That really surprised me as I expected the slow speed to result from traffic congestion and I know most Internet traffic shows strong time of day variations. So next I looked at the Pingtest results averaging tests by the time of day in order to get another hour by hour view. Among other things, Pingtest measures jitter and packet loss. When links become congested the first indication is usually increased jitter. As congestion become significant, you begin to see packet loss.
It's not dramatic, but there appears to be an increase in jitter during the evening hours.
Packet loss is usually zero, but when it does occur, it drives the hourly averages to greater than zero and this graph shows a distinct increase during the evenings.
Conclusions
Unfortunately, from the data and analysis so far, I can't pinpoint what is happening. The performance of my FiOS service is highly variable, but the problem or problems are not restricted to specific times of day, e.g. high traffic intervals.
If anyone is interested in the measurements I've made so far, just ask and I'll shoot you the spreadsheets containing the raw data. Meanwhile, I'll be considering what new measurements might shed light on what's going on.
