Getting low speeds reported

[FiberGuy]

I have 50/50 rated Active Gig Ethernet with a ping to Level 3 Chicago that has a min of 10ms and an average of 11ms.

 

I typically don't have issues downloading, but for some reason, I am getting something like 3-4Mb/s from these tests.

 

 

 

This is what I'm used to. The spike going down was my HD thrashing from swap.

[mudmanc4]

Try different testing servers, use the CentralUS server as the control, move to washington DC and any other official testmy.net servers. 

 

Then you can move to publicly hosted servers for testing routes and providers. 

[FiberGuy]

It seems there is some sort of artificial limitation, results are too consistent. I tired Chrome, FireFox, and IE, all up-to-date. Not any huge difference.

 

Does Win7 need TCP settings tweaks? I get 114MB/s on my LAN via SMB file transfers, and was just getting 50Mb/s from a single download for several minutes strait on Steam(Civ5 Gold Upgrade was only $5).

 

[FiberGuy]

It seems the servers being used are from SL(SoftLayer). My ISP's upstream provider is L3(Level3). SL has 80Gb/s of peering with L3 in Dallas, but I am in the Midwest.

 

It seems SL only has 10Gb of peering with L3 in Chicago.  What happening is that my tracert goes from my ISP, to L3 Chicago, to SL Chicago, to SL Dallas. Now, SoftLayer only has a total of 20Gb of their own private bandwidth inter-connecting their datacenters.

 

When I tracert SL Dallas, the jump from SL Chic to SL Dal is a 24ms hop, while L3 Chic to L3 Dal is only 19ms. Since they take nearly the same path and the speed of light is quite fast, my conclusion is that SL's Chic to Dal link is a bit congested.

 

It seems to me that my speedtest issues are that L3 seems to be hot-handoff routing.

 

This is the best conclusion I can come to with my limit Google-Foo abilities.

[ybnrmalatall]

It seems the servers being used are from SL(SoftLayer). My ISP's upstream provider is L3(Level3). SL has 80Gb/s of peering with L3 in Dallas, but I am in the Midwest.

 

It seems SL only has 10Gb of peering with L3 in Chicago.  What happening is that my tracert goes from my ISP, to L3 Chicago, to SL Chicago, to SL Dallas. Now, SoftLayer only has a total of 20Gb of their own private bandwidth inter-connecting their datacenters.

 

When I tracert SL Dallas, the jump from SL Chic to SL Dal is a 24ms hop, while L3 Chic to L3 Dal is only 19ms. Since they take nearly the same path and the speed of light is quite fast, my conclusion is that SL's Chic to Dal link is a bit congested.

 

It seems to me that my speedtest issues are that L3 seems to be hot-handoff routing.

 

This is the best conclusion I can come to with my limit Google-Foo abilities.

Adding this to my double like

Google-Foo had me rolling 

Edited June 19, 2013 by ybnrmalatall

[FiberGuy]

Using file transfers and looking at my NIC bandwidth, I get better speeds to these places than the speed test servers.

 

Japan, Taiwan, China, India, Australia, Singapore, South Africa, North-East Africa, Arabia, South America, Central America.

 

It seems this site needs a Chicago presence. I can appreciate the effort put into making a good speed test, but the "TestMy.net Bandwidth Test Legitimacy" section of the site makes it sound like these tests are nearly infallible and if there is a speed issue, it's your ISP.

 

In my case, it is not my ISP, but the hosting provider which this service is running on. There are always corner cases, but it's a bit haughty to say "Our servers are also configured and tested to maintain full quality of service for thousands of miles." when they bottle-neck in some corner cases is the internal network of the Speed Test service.

 

It's not just this site. I get similar issues with SpeedTest.Net and a few other sites. They seem to connect with Level 3 in the local region, then route their own inter-regional traffic.

 

While my speed tests tend to be low, my bandwidth works where it counts. I get great speeds to real services. Kind of an inverse situation from my last ISP, where I got great speed tests and relatively poor real world speeds.