Speed problem

[haakon]

Hi,

I have two RB1000 connected to eachother, with 1Gb ethernet connection.

When I try to do a bandwidth test, both TX and RX goes up to about 100mbit, before TX speed suddenly goes down to 15~mbit, while RX continues up to ~970Mbit.
Anyone knows why this happen?

It’s no magic on the interfaces, they are not part of any bridges, just plain interfaces. None of the devices has any firewall enabled.

I tried the other way too, and the RX and TX actually swaps, so seems like consequently TX on one box, or RX on the other is having problem?

Switching cables did not have any impact.

[Chupaka]

CPU usage? TCP test?..

[normis]

you generate the random data, send the data, gather the data, and display the data all on the same device. what did you expect? in normal scenario, the user generates, sends and receives the data, no the router. router only routes. modify your test so that the router doesn’t have to do all this stuff and kill it’s own CPU

[haakon]

Shouldn’t both routers work equally to generate random data?

This is unclear:

Bandwidth tester, does the “bandwith test server” just reply the data you transmit to it, or does it generate data itself?

Router A Router B
GENERATE ----------> RECEIVE
RECEIVE <----------- GENERATE

or

Router A Router B
GENERATE ------------> RECEIVE
RECEIVE <------------ SEND BACK

If the first case is correct, then it shouldn’t “die” in just one direction. Because they would be working equally hard to generate, transmit and display data.

AND, I have tried running the test (displaying) from both routers, and it’s the same direction that dies. Like it would if there was problems with one pair in the tp cable, or the network device. But I have tried different ports and cables. With same result.

I have set up one more router now. So I will try to check speed A->B->C.

[haakon]

This is the result for A->B->C.

Router B is just set to Soft Bridging A and C with no fw enabled.

Is this within expected results for a soft bridge with 1500byte packets?

With TCP, i get ~139mbit both ways.

[normis]

This is what we get, using a specialized network traffic load simulator:
http://www.routerboard.com/pdf/tests_april_2009.pdf

[haakon]

In the A<->B<->C test, the CPU status is as follows:

A = 40% CPU (Bandwith tester)
B = 40% CPU (Bridge)
C = 99% CPU (Btest server)

Average 515mbit in and 360mbit out

Testing C<->B<->A:

C = 67% (Bandwith tester)
B = 41% (Bridge)
A = 35% (Btest server)

Average 400mbit in and 522mbit out


I still think this is weird. Disagree?

[haakon]

I saw that I had forgotten to upgrade router A to RouterOS 4.0.

Now that I did that, instead of ~500mbit both ways I’m seeing the 900Mbit out 25mbit in issue from A->B->C !

So something big in either the bridge or the bandwidth tester has happened between the OS shipped with the routers, and the latest 4.0.

A = 19% CPU
B = 44% CPU
C = 100% CPU

Now, when I test from C i get the following, more concequent data:

C = 100% CPU
B = 44% CPU
A = 19% CPU.

Why is it that, both A and C routers are clean, no config, just firmware upgraded. Have so different cpu usage?

I begin to wonder if router C is broken.

[normis]

make that x2 because you seem to only look at one direction speed. look at your graphs.

[haakon]

@normis

make that x2 because you seem to only look at one direction speed. look at your graphs.

What do you mean only look at one direction speed?
I’m sending traffic both ways, with the “both” function.

And what should i make x2?

The graphs are saying that something is wrong with router C

Router A->B gives me 900 out, 671 in.
Router C->B gives me 22 out, 972 in