Huge speed disparity between windows/linux (bad for linux)

Errsta_Fonzarelli · 03-04-2006, 10:10 PM

Delete Delete

macemoneta · 03-04-2006, 11:47 PM

"Broadband speed tests" are worthless. They send a relatively small amount of data from a single host, over a single route.

If I take one of these tests repeatedly over a 15 minute interval, there's about a 300% variation in the reported speed.

To properly test download speed, traceroute several high-bandwitdth (100Mb+) sites, and select 4 or 5 with the least common routes (aside from the connection through your ISP to your machine). Start a large (minimum of 500MB) download from each, at the same time. Monitor your aggregate bandwidth (gkrellm or similar interface monitor).

Yup, it's a lot of effort to get a real number.

Errsta_Fonzarelli · 03-04-2006, 11:52 PM

Delete Delete

peter_robb · 03-05-2006, 04:45 AM

Also remember that M$ includes Qos scheduling which native linux doesn't. You need to add that yourself. The usual difference will be buffer congestion somewhere in between..

Have a read of http://www.tldp.org/HOWTO/Adv-Routing-HOWTO/index.html
particularly chapter 15.8 http://www.tldp.org/HOWTO/Adv-Routin...timate-tc.html

macemoneta · 03-05-2006, 09:30 AM

QoS only applies when there are multiple streams contending for bandwidth. Was there a high bandwidth activity going on on the Linux machine at the time you were attempting the test?

Errsta_Fonzarelli · 03-05-2006, 12:39 PM

Delete Delete

macemoneta · 03-05-2006, 01:33 PM

Then perhaps what you are measuring is an artifact of the java interpreter in each environment, and not a reflection of the actual bandwidth utilization.

Errsta_Fonzarelli · 03-05-2006, 01:39 PM

Quote:

Originally Posted by macemoneta

Then perhaps what you are measuring is an artifact of the java interpreter in each environment, and not a reflection of the actual bandwidth utilization.

I was kind of thinking something along those lines...when in doubt, blame java

I'm looking for a non-java tester to test that theory.

Intimidator · 03-05-2006, 02:10 PM

I think this has something to do with IPv6 module enabled by default in the kernel.

Disabling IPV6

Errsta_Fonzarelli · 03-05-2006, 02:53 PM

Delete Delete

davcefai · 03-05-2006, 03:18 PM

Have you checked the MTU size?

Errsta_Fonzarelli · 03-05-2006, 03:30 PM

Delete Delete

peter_robb · 03-05-2006, 03:31 PM

An easy raw tester is wget..

Set it up to download a big iso.
Then add rate switches from something slow in steps upward until it flattens out.
Watch the rate indicator until it stabilises then Ctrl C to interrupt and try another rate..
eg
wget -c --limit-rate=20K http://server/*.iso

macemoneta · 03-05-2006, 03:42 PM

I ran the speed test you listed, and it reported 4Mb. I've confirmed that it is actually 8Mb using the method I described above. Since the "test" runs for only a couple of seconds, it has no value. In my case, it's report has a 100% error skew (at least).

Update: I ran a "test" at another site, and it reported 11Mb/s. Considering that my cable modem has a 10Mb/s interface, that really is impressive.