Disabling ipv6

Chris Adams linux at cmadams.net
Thu Jul 11 19:47:37 UTC 2013


Once upon a time, Richard Sewill <rsewill at gmail.com> said:
> I tried ping and ping6 anyway.  This is NOT on an idle network.

Since ICMP and ICMPv6 are low-priority, the data is not very useful.
Also, since latency is only one component of throughput (and most
communications are not particularly sensitive to latency less than about
200ms, except for issues like bufferbloat), this really doesn't mean
much.

However, since we're going for anecdotal evidence, this is on an
otherwise idle system on an uncongested link (and not using a tunnel):

$ ping -c5 www.google.com
PING www.google.com (74.125.26.106) 56(84) bytes of data.
64 bytes from vh-in-f106.1e100.net (74.125.26.106): icmp_seq=1 ttl=40 time=45.2 ms
64 bytes from vh-in-f106.1e100.net (74.125.26.106): icmp_seq=2 ttl=40 time=45.2 ms
64 bytes from vh-in-f106.1e100.net (74.125.26.106): icmp_seq=3 ttl=40 time=45.3 ms
64 bytes from vh-in-f106.1e100.net (74.125.26.106): icmp_seq=4 ttl=40 time=45.7 ms
64 bytes from vh-in-f106.1e100.net (74.125.26.106): icmp_seq=5 ttl=40 time=45.5 ms

--- www.google.com ping statistics ---
5 packets transmitted, 5 received, 0% packet loss, time 4052ms
rtt min/avg/max/mdev = 45.238/45.443/45.796/0.244 ms
$ ping6 -c5 www.google.com
PING www.google.com(vh-in-x6a.1e100.net) 56 data bytes
64 bytes from vh-in-x6a.1e100.net: icmp_seq=1 ttl=55 time=24.8 ms
64 bytes from vh-in-x6a.1e100.net: icmp_seq=2 ttl=55 time=24.8 ms
64 bytes from vh-in-x6a.1e100.net: icmp_seq=3 ttl=55 time=24.8 ms
64 bytes from vh-in-x6a.1e100.net: icmp_seq=4 ttl=55 time=24.9 ms
64 bytes from vh-in-x6a.1e100.net: icmp_seq=5 ttl=55 time=24.9 ms

--- www.google.com ping statistics ---
5 packets transmitted, 5 received, 0% packet loss, time 4033ms
rtt min/avg/max/mdev = 24.819/24.878/24.909/0.202 ms

> There remains a performance penalty when using IPv6.

No, there is possibly a performance issue with your ISP.  Have you
reported the problem?

> Given the ISP is handing out 6to4 tunneling, I still think the ISP support
> is "sort of not there".

Lots of ISPs will probably use last-hop tunnels for a while, because a
lot of the last-hop gear is old and doesn't properly support IPv6.
Eventually that gear will be replaced, but in the interim, they'll
install tunnel servers alongside the last-hop gear.  It is possible your
ISP doesn't have the tunnel server near your last-hop and is taking a
sub-optimal path.

However, similar kinds of sub-optimal routing happen with IPv4 all the
time, especially once MPLS comes in to play.

> I am guessing, please correct me if I am wrong, IPv4 will be used in
> preference to IPv6, when both are available.

No, when both are available, IPv6 takes precedence (in general for
modern applications that don't override the precedence); this is spelled
out in several RFCs (can't recall the numbers).  I think there is a
global way to override this (maybe /etc/gai.conf can do it?).

-- 
Chris Adams <linux at cmadams.net>


More information about the users mailing list