Skip to content

Latest commit

 

History

History
143 lines (109 loc) · 3.85 KB

README.md

File metadata and controls

143 lines (109 loc) · 3.85 KB

Xcluster/ovl - mpls

Tests and experiments with MPLS. This is a complement to ovl/srv6.

MPLS itself is really simple. The marketed benefits of MPLS, such as "50ms failover time", "Improve User Experience", etc. are not properties of MPLS but how MPLS is setup, i.e. the control plane. However MPLS makes these things possible.

Network

The diamond network topology is used;

Manual test

Check the help printout for ./mpls.sh for automatic tests.

Start cluster

./mpls.sh test start > $log

Now we have no connectivity vms<->testers. Our job is to setup this using MPLS.

# On vm-001
ping 192.168.2.221    # Doesn't work
# On vm-201
ip route add 192.168.2.0/24 encap mpls 201 via inet6 $PREFIX:192.168.3.203
ip route add $PREFIX:192.168.2.0/120 encap mpls 201 via inet6 $PREFIX:192.168.3.203
ip -f mpls route add 203 dev lo
# On vm-202
ip route add 192.168.1.0/24 encap mpls 202 via inet6 $PREFIX:192.168.5.203
ip route add $PREFIX:192.168.1.0/120 encap mpls 202 via inet6 $PREFIX:192.168.5.203
ip -f mpls route add 203 dev lo
# On vm-203
ip -f mpls route add 201 as 203 via inet6 $PREFIX:192.168.5.202
ip -f mpls route add 202 as 203 via inet6 $PREFIX:192.168.3.201
# On vm-001
ping 192.168.2.221    # Works!
ping $PREFIX:192.168.2.221

I direct packets to lo for decapsulation. Is there a better way?

You may capture traffic to see what's happening;

# On host
./mpls.sh test --no-stop > $log
xc tcpdump --start 203 eth1
xc tcpdump --start 203 eth2
# On vm-001
ping -c1 192.168.2.221
# On host
xc tcpdump --get 203 eth1 eth2
xc stop

Segment Routing

We have looked as normal MPLS, now we continue with MPLS-SR. Segment Routing is about source routing so we will add multiple labels at the starting endpoint. Example;

ip route add 192.168.2.221/32 encap mpls 203/202/221 via inet6 $PREFIX:192.168.1.201

The MPLS-SR aware routers must be configured to consume the top label and forward the packet. Example;

ip -f mpls route add 203 via inet6 $PREFIX:192.168.3.203

Manual test

In this setup all routers are MPLS-SR aware and MPLS is setup on all nodes.

./mpls.sh test --nvm=1 start_full_mpls > $log

We must setup the segments on the endpoints, i.e. the vms and testers (only vm-221). Note;

#define MPLS_LABEL_FIRST_UNRESERVED   16 /* RFC3032 */

So we offset the labels on the vms by 100, so the label for vm-001 becomes 101, etc.

# On vm-001
ip route add 192.168.2.221/32 encap mpls 203/202/221 via inet6 $PREFIX:192.168.1.201
# On vm-201
ip -f mpls route add 203 via inet6 $PREFIX:192.168.3.203
ip -f mpls route add 101 via inet6 $PREFIX:192.168.1.1
# On vm-203
ip -f mpls route add 201 via inet6 $PREFIX:192.168.3.201
ip -f mpls route add 202 via inet6 $PREFIX:192.168.5.202
# On vm-202
ip -f mpls route add 203 via inet6 $PREFIX:192.168.5.203
ip -f mpls route add 221 via inet6 $PREFIX:192.168.2.221
# On vm-221
ip route add 192.168.1.1/32 encap mpls 203/201/101 via inet6 $PREFIX:192.168.2.202
ping 192.168.1.1   # Works

Note that the entire network is pure IPv6 but we carry IPv4.

Again to capture traffic is interresting;

# On host
./mpls.sh test --no-stop basic_sr > $log
xc tcpdump --start 203 eth1
xc tcpdump --start 203 eth2
# On vm-221
ping -c1 192.168.1.1
# On host
xc tcpdump --get 203 eth1 eth2
xc stop

References