[LARTC] netem bug?

[Catalin(ux aka Dino) BOIE]

> Hi all,
>
> I'm running some tests with netem and I noticed some strange behaviour that 
> looks like a bug: I'm pinging another machine and adding delay with netem. 
> When I tell netem to give me a 10ms delay, it works fine. The problem is that 
> when I ask for a 11ms delay, it gives me 20ms! It happens for any value 
> between 11ms an 20ms, and it repeats for values over 20ms, now rounding up to 
> 30ms.
>
> Here is an example:
>
> # ping x.x.x.x
> PING x.x.x.x (x.x.x.x) 56(84) bytes of data.
> 64 bytes from x.x.x.x: icmp_seq=1 ttl=64 time=0.197 ms
> 64 bytes from x.x.x.x: icmp_seq=2 ttl=64 time=0.100 ms
> 64 bytes from x.x.x.x: icmp_seq=3 ttl=64 time=0.081 ms
> 64 bytes from x.x.x.x: icmp_seq=4 ttl=64 time=0.099 ms
>
> # tc qdisc add dev eth0 root netem delay 10ms
> # ping x.x.x.x
> PING x.x.x.x (x.x.x.x) 56(84) bytes of data.
> 64 bytes from x.x.x.x: icmp_seq=1 ttl=64 time=1.11 ms
> 64 bytes from x.x.x.x: icmp_seq=2 ttl=64 time=9.04 ms
> 64 bytes from x.x.x.x: icmp_seq=3 ttl=64 time=9.24 ms
> 64 bytes from x.x.x.x: icmp_seq=4 ttl=64 time=10.0 ms
>
>
> # tc qdisc change dev eth0 root netem delay 11ms
> # ping x.x.x.x
> PING x.x.x.x (x.x.x.x) 56(84) bytes of data.
> 64 bytes from x.x.x.x: icmp_seq=1 ttl=64 time=13.1 ms
> 64 bytes from x.x.x.x: icmp_seq=2 ttl=64 time=20.0 ms
> 64 bytes from x.x.x.x: icmp_seq=3 ttl=64 time=20.0 ms
> 64 bytes from x.x.x.x: icmp_seq=4 ttl=64 time=20.0 ms
>
> # tc qdisc change dev eth0 root netem delay 23ms
> # ping x.x.x.x
> PING x.x.x.x (x.x.x.x) 56(84) bytes of data.
> 64 bytes from x.x.x.x: icmp_seq=1 ttl=64 time=22.2 ms
> 64 bytes from x.x.x.x: icmp_seq=2 ttl=64 time=29.8 ms
> 64 bytes from x.x.x.x: icmp_seq=3 ttl=64 time=29.1 ms
> 64 bytes from x.x.x.x: icmp_seq=4 ttl=64 time=30.0 ms
>
> # uname -a
> Linux yyy 2.4.27-1-386 #1 Fri Sep 3 06:24:46 UTC 2004 i686 GNU/Linux
>
>
> Is this a known bug? Is there any fix? I also noticed that the first ping has 
> always a smaller value. Can someone explain that?

The "problem" is the 2.4 kernel because HZ is 100.
Ping is not rescheduled quickly enough to measure correctly the rtt.

On my 2.6 kernel (HZ=1000) it work almost perfect.

Hope it helps.

>
> Thanks,
>
> Pedro.
>
> _______________________________________________
> LARTC mailing list / LARTC@mailman.ds9a.nl
> http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
>

---
Catalin(ux aka Dino) BOIE
catab at deuroconsult.ro
http://kernel.umbrella.ro/