Gigabit Service with Supermicro A1SRi-2758f

mifronte

Build Information

I followed this thread for my build. I had to set the kern.ipc.nmbclusters=512000 to get rid of the Igb3: Could not setup receive structures error. I also enabled trim support for the SSD.

So I recently had fiber gigabit service installed (fiber to the modem) and have been doing speed testing. Here is a couple recent test results:

| | |

During my testings, I've noticed that any test sites utilizing HTML5 instead of Flash (testmy.net, speedof.me), gave very erratic and low results (average around 65Mbps). Between tests, the speed will vary between 45-200Mbps. Are the HTML5 based tests not as accurate?

pfcode

Forget these sites, they aren't capable of doing 1gbps test, even got wrong results for my 250/20.

mifronte

@pfcode:

Forget these sites, they aren't capable of doing 1gbps test, even got wrong results for my 250/20.

Just to clarify, you are speaking about the HTML5 sites or all test sites?

pfcode

testmy.net, speedof.me. not all sites, speedtest.net, http://www.dslreports.com/speedtest are working fine.

mifronte

I tried the DSLReports speed test and I am getting an F on the BufferBloat. I have a symetrical gigabit fiber service. Looks like the BufferBloat is occurring on the download test. Does this mean I have to do some traffic shaping on the LAN interface?

I am reading about CODELQ but can someone suggest a rule of thumb setting for the WAN and LAN interface?

Guest

Between tests, the speed will vary between 45-200Mbps. Are the HTML5 based tests not as accurate?

They all have one criteria that can not be wiped away, by using this tests for getting a result you
will even and only get "something around" numbers about your Internet connection:

The quality of cabling infrastructure of your household and during the city you are living in
How many peoples are using this test server at the same time?
How long is the distance from yours to the test server?
Is this test server owner paid by some or more ISPs for good throughput numbers shown to their customers?
How many customers of your ISP are sharing together one router on the ISP side?
How much is going in the entire internet during this test? Saturday morning were all a sleeping or car washing,
or shopping, or cutting the green or…... There will be less traffic on the entire Internet and you will be getting
probably better numbers as Friday evening, or am I wrong with this?

Just to clarify, you are speaking about the HTML5 sites or all test sites?

HTML5 or Ajax based is not the real thing in my eyes, the reaction time is relevant to this
numbers what does it mean, having on the test side a HTML5 page loading and the freaking
server is overloaded?