Understanding single thread sluggishness

[nkruis]

We have a dedicated 20/20 connection on CenturyLink fiber for our business. Of course, they only care about their own OOKLA tests, but we've been finding that our single thread TMN speeds are closer to 3-4Mbps. They claim that there should be nothing limiting a single thread download, and chalk it up to "normal internet variations". Regardless of when we test, we always see the same limitation at about 3-4Mbps for single thread TMN tests. Multi-thread tests are better (15-18Mbps). Because the OOKLA tests come in at the "right speed", they claim that this isn't a problem.

 

My question is: What should we expect? Is this specific to CenturyLink, or is this standard behavior for ISPs? Should we be switching providers?

 

https://testmy.net/compID/615662894136

Edited January 20, 2016 by CA3LE
added live graph

[mudmanc4]

I would expect a bit more from that connection.

At the same time, a leased line, if this is simply as stated a dedicated fiber line to your ISP, and not connecting corporate offices, that if you tested say 5, or 25 machines at the same time, the results could be reflected much the same as your seeing with one.

[In other words, each workstation might achieve the same ~5Mbps at the same time another or others}

 

Are you in a position to simultaneously test any multiple of workstations to verify?

[nkruis]

UPDATE: One of the issues was our test server. Selecting a closer server gave us a better single test result (around 8-9Mbps). Better than 3-4Mbps, but still only half of what we're paying for.

 

I just tested three machines at the same time with no other significant Network traffic. Our results: 7.3, 5.7, and 4.7Mbps for a combined 17.7 down. That seems reasonable, but I still have questions about our single thread performance:

 

- What is actually limiting our single thread downloads? Is the ISP throttling single connections? (They claim that there is no reason a single thread should be slower, but they also don't care that it is because they only look at the OOKLA test).

- If we upgrade our service, to say 40/40, would we expect to see our single thread speeds to also double?

https://testmy.net/quickstats/nkruis

Edited January 20, 2016 by mudmanc4
added charting

[mudmanc4]

The test they are using is to verify the end to end connection, within their internal networks. It has no other purpose. If hosted on [or within] their networks.

 

Just for fun (it's all fun though isn't it) start a download of debian [http://cdimage.debian.org/debian-cd/8.2.0/amd64/iso-dvd/debian-8.2.0-amd64-DVD-1.iso] which is 3.7 gigs in IE and one in Firefox or Chrome on one machine.

 

Monitor whatever you have which can print or render live throughput.

[nkruis]

I tried downloading 5 instances of the debian disk image at once: 3 on Chrome, 1 on Firefox, and 1 on Safari. Running activity monitor, my computer showed consistently between 200-500KBps (1.6-4Mbps).

[mudmanc4]

3 hours ago, nkruis said:

I tried downloading 5 instances of the debian disk image at once: 3 on Chrome, 1 on Firefox, and 1 on Safari. Running activity monitor, my computer showed consistently between 200-500KBps (1.6-4Mbps).

Going out on a mental limb here, what happens when this is done on three separate machines locally.

I mean to say, one instance of the download per workstation.

Might sound strange or leading to inconclusive results, maybe, just my way of analyzing. 

[nkruis]

3 downloads (Safari, Chrome, and Firefox) on 3 machines. Overall, the downloads were typically between 100-250KBps with occasional spikes to 400-500KBps or some machines. In total, across three machines, I'd say we were pretty steady at 600KBps (4.8Mbps) total.

 

There doesn't seem to be much of a difference between one or multiple instances of the download.

[mudmanc4]

I may have missed this somewhere, what are you using to manage traffic / IP's ect. Router / firewall.

 

I ask to see if there could be some QOS / traffic shaping going on, in any sense.

 

[nkruis]

We are using a Juniper SRX 240 Gateway and an Apple Airport Extreme WiFi router. The low speeds seem to occur even with a direct ethernet connection.

[mudmanc4]

8 minutes ago, nkruis said:

We are using a Juniper SRX 240 Gateway and an Apple Airport Extreme WiFi router. The low speeds seem to occur even with a direct ethernet connection.

Ah, so with or without the switch, results are ~the same?

Even using different workstations / laptops what have you?

 

If not, run:

show policer

 

[nkruis]

Admittedly, we don't have the experience in-house to login to the Junos operating system.

[mudmanc4]

4 minutes ago, nkruis said:

Admittedly, we don't have the experience in-house to login to the Junos operating system.

Everything you have said, mainly the section on removing the switch from the equation produces the same results, I would continue submitting tickets to the ISP, they will eventually respond and send a tech.

 

Of course they will not be responsible for administering the local network, but with enough perseverance they should responsibly cave, and get someone out there.

[nkruis]

Reviving the old discussion here. I've done more testing at the switch and I'm getting the expected results there. I've identified the router as the problem. The Apple Airport Extreme is limiting the single thread downloads (e.g., through TMN) are limited, while Speedtest.net tests show the full speeds we're paying for. Any ideas how we can configure the router to fix this behavior?

[mudmanc4]

13 hours ago, nkruis said:

Reviving the old discussion here. I've done more testing at the switch and I'm getting the expected results there. I've identified the router as the problem. The Apple Airport Extreme is limiting the single thread downloads (e.g., through TMN) are limited, while Speedtest.net tests show the full speeds we're paying for. Any ideas how we can configure the router to fix this behavior?

Click multi threaded and explain what the results are, be sure to use several servers.