ISP marking ipsec traffic based on certificate, how is this possible?

[Mark Zimmer <sgi () tango lu>]

Hello list,

 I have a site-to-site ipsec vpn with strongswan. It was working well
 for 5-6 months then a day ago I have noticed something strange, that
 from Site-A to Site-B (tunnel mode) only the upload bandwidth is 
capped
 down to 20-30kbit/s inside the VPN.
 I have tried various apps like ftp, scp on different ports it was the
 same result. I also ran speedtest/wget on both endpoints just to make
 sure that not the entire connection of those networks are capped.

 Since outside parties cannot see anything from what's going on inside
 the tunnel, first I was thinking that they started limiting the 
traffic
 based on port (4500 udp) or based on protocol (ESP), that is easy to 
do.

 In older versions of strongswan it's not possible to change the charon
 nat port (probably wouldn't work anyway since most of the traffic 
should
 be ESP (protocol 50)).
 I have restarted the strongswan daemon on both endpoints multiple 
times
 it did not change the situation (the bandwidth limiting was still 
present).

 So my last idea was to make new vpn certificates. For my biggest
 surprise with the new certificates the capping was gone and the
 bandwidth went back to normal. I hope I don't have to put the old 
certs
 back from backup just to make a point.

 One of the ISPs must started tagging the ipsec traffic based on the
 certificate and then do traffic shaping (QoS) on it to throttle down 
the
 bandwidth. How is this even possible? I was thinking that an ipsec
 connection is encrypted and random from the beginning. How can they
 define a pattern to their whatever device to be able to mark this
 specific traffic?
 Is there a part at the beginning of the connection sequence which is
 always the same with using the same certificate?

 Do I have to worry about here that my vpn keys got compromised?

 Anybody ever experienced this?

 Thanks!