Discrepancies in Internet Speed Tests

[keylimesoda]

I'm paying for 50 Mbps from Comcast.

I've run the following speed tests, three times a day, for the last week. These are the average results:

1. http://www.speedtest.net/ (16/7)
2. https://www.speakeasy.net/speedtest/ (16/6)
3. http://speedtest.comcast.net/ (54/7)

Additionally, when I do Steam downloads, they cap off at ~2MBps (bytes), which backs up the 16Mbps speeds from the other tests.

So, how do I drive the discussion with Comcast that I'm paying 50Mbps prices for 16MBps speeds? Or am I missing something on these speedtests?

 

[CaptainSTX]

Are you testing using an Ethernet connection from your PC to Comcast's modem? If you are using your own router and WiFi you have added to many variables to necessarily blame Comcast. Also testing using phones and tablets may not give you a true picture of the speed Comcast is or isn't delivering.

You need to be running speed tests to speed test servers nearby?

Have you tested the speeds using multiple PCs? Have you tested you PC on some other network and what speeds did you get there?

You need to isolate the problem by eliminating as much as possible your equipment as the possible bottle neck. Once you have done that you will have the facts to have a meaningful conversation.

 

[keylimesoda]

Good questions! All tests were run in the following configuration:

HP Gen6 Proliant Server->1GB Switch->PFSense router->Motorola Surfboard (SB6121)

I tested against servers from Seattle to Houston, and results were similar (huge on Comcast, slower on others).

Additionally, internet speed show the same for other hardwired PCs on the network, both in synthetic tests and in actual practice (Steam downloads, linux distro downloads, etc.)

What's odd is not just the poor performance--it's that the Comcast test *doesn't* show the poor performance.

Should I eliminate the switch and router before contacting Comcast? They seem unlikely failure candidates, but I could understand if they're going to get picky about that.

 

[Nullity]

Good questions! All tests were run in the following configuration:

HP Gen6 Proliant Server->1GB Switch->PFSense router->Motorola Surfboard (SB6121)

I tested against servers from Seattle to Houston, and results were similar (huge on Comcast, slower on others).

Additionally, internet speed show the same for other hardwired PCs on the network, both in synthetic tests and in actual practice (Steam downloads, linux distro downloads, etc.)

What's odd is not just the poor performance--it's that the Comcast test *doesn't* show the poor performance.

Should I eliminate the switch and router before contacting Comcast? They seem unlikely failure candidates, but I could understand if they're going to get picky about that.

Since you are able to achieve 50mbit consistently with Comcast's servers, the problem seems to be outside the house, not inside.

Since you are running pfSense, the RRD graphs should show you your bandwidth over time. These graphs should be able to give you much more real-world information than some speedtests. Though, it does sound like you are NOT getting the speed you paid for, but I cannot be sure of that.

 

[theonlyski]

Since you are able to achieve 50mbit consistently with Comcast's servers, the problem seems to be outside the house, not inside.

Since you are running pfSense, the RRD graphs should show you your bandwidth over time. These graphs should be able to give you much more real-world information than some speedtests. Though, it does sound like you are NOT getting the speed you paid for, but I cannot be sure of that.

I've never known any ISP to say that you will get your advertised speeds outside of their networks.

That being said, it might be worth calling, but chances are, there's not much they're going to do.

 

[htismaqe]

I've seen this before. It's very possible that the bottleneck exists on the peering network between Comcast and whatever provider is providing transit to other networks.

Unless you really know what you're doing and have access to BGP peering information and other tools, proving this (as a residential customer) is nearly impossible.

You best hope is to suggest it as a possible problem and hope that Comcast takes it seriously.

 

[Nullity]

I've never known any ISP to say that you will get your advertised speeds outside of their networks.

That being said, it might be worth calling, but chances are, there's not much they're going to do.

I was hesitant about contacting my ISP, assuming they would not be able to do anything or that they would simply ignore me. When I finally contacted them and told them my situation, they apologized for making a mistake with the BRAS (ADSL) configuration. Once I reconnected, my connection lost all throughput jitter and the throughput increased from ~4mbit to 7mbit (I pay for 6mbit). From start to finish, it was less than an hour to get my connection fixed.

I felt pretty stupid for not contacting them sooner...


Though, make sure the error is on their side before you contact them. I'm sure ISP's get plenty of generic "my internet is not fast enough" complaints from oblivious consumers who don't even know what a megabit is.

 

[keylimesoda]

Just got off the phone with Comcast.

They acknowledged that they'd never correctly added my new SB6121 to my account (I installed it a year ago). They weren't sure how I was getting internet

Unfortunately, even after that, the speeds haven't changed. Good news is that they see the slow speeds on their side and they're sending a tech out on-premises in two days to troubleshoot the issue.

Customer service rep was solid--it's obviously been a focus for them lately. I'll update back here once things get resolved.

 

[keylimesoda]

Should've connected directly from modem to PC before I called Comcast.

It turns out the issue was in my router <facepalm>

I'm running PFSense, and had configured the QOS back when I was running at 20Mbps. This meant that the QOS limited the system throughput.

So, disabling QOS, I saw full speed on all my speed tests and activities.

Oddly, what this still doesn't explain is how the Comcast test was fast when the rest of the tests were impacted by QOS. The only thing I can figure is that the Comcast test was running on a different port/protocol?

Now I need to go take some lessons on how to configure QOS such that I don't take a hit on my download speeds.

EDIT: Lessons done. Configured at a higher connection speed and started using priq instead of hfsc. Now I'm seeing no loss of throughput on any of my devices. Sweetness.

 

[Klueless]

I'm running PFSense, and had configured the QOS back when I was running at 20Mbps. This meant that the QOS limited the system throughput.

So, disabling QOS, I saw full speed on all my speed tests and activities.

Oddly, what this still doesn't explain is how the Comcast test was fast when the rest of the tests were impacted by QOS?

Seems like a good question. I also use bandwidth limiters alongside QoS. Seems to work. "DSL Reports" bufferbloat improved from F to A when I implemented bandwidth limiters. Web feels perkier when kids are streaming and gaming.

Most speed tests seem to reflect the throttled speed. Intriguingly "Speed of Me" and "nperf" reflect raw (unthrottled) speed. I'm also curious as to why?

Regardless, it's useful. Unknown to me my ISP actually increased my speed. If it wasn't for these "renegade" speed tests I would have never known (to increase my bandwidth limiters).

 

[sfx2000]

I'm running PFSense, and had configured the QOS back when I was running at 20Mbps. This meant that the QOS limited the system throughput.

So, disabling QOS, I saw full speed on all my speed tests and activities.

Serves as a good reminder - pfSense traffic shapers limit the up and down link - so if upgrading to a higher tier of service (or if the ISP bumps up the current tier subscribed to), one has to update the shaper bandwidths are well - otherwise, like OP mentions...