Last week of February 2015, Franck Simon, managing director at the internet exchange point France-IX, gave three presentations about the Best Common Practices for the IXPs.
This first part of the trilogy focuses on the practices used by the IXP.
2. BCP for IXPs - PoPs
• Neutral
coloca+on
(ideally
datacenters)
for
PoPs
• Layer-‐2
infrastructure
(either
na+ve
Ethernet
or
VPLS
based)
• Resiliency
and
scalability
:
– Equipment
redundancy
or
spare
equipment
for
PoPs
– When
mul+-‐PoPs
architecture
:
edge
PoPs
&
core
PoPs
vs
full-‐mesh
design
– Use
of
passive
WDM
for
PoPs
interconnec+ons
(when
appropriate
and
possible)
2
3. BCP for IXPs - NOC
• NOC
(outsourced
or
run
internally)
and
technical
support
3
4. BCP for IXPs – Peering
LAN
• Same
VLAN
for
Unicast
IPv4/IPv6:
– IPv4
&
IPv6
• Specific
VLAN
for
Mul+cast
(if
needed)
• Specific
VLANs
for
private
peering
(closed
user
groups)
4
5. BCP for IXPs – Technical
requirements
• Provide
na+ve
layer-‐2
(or
fully
transparent
layer-‐2
infrastructure)
• Block
spanning-‐tree
on
all
members
ports
• Allow
only
authorized
EtherTypes
(IPv4,
IPv6,
ARP)
• Limit
MAC
learning
(one
MAC/port)
and
apply
MAC
filtering
• Use
storm
control
(broadcast/mul+cast)
or
apply
rate-‐limi+ng
filters
to
limit
broadcast
• Allow
LACP
• Quaran+ne
VLAN
(mainly
for
new
members)
=>
Don’t
connect
a
new
member
on
the
IXP
without
running
a
strict
valida+on
procedure
5
6. BCP for IXPs – Main
services
• NTP
service
• DNS
servers,
DNS
root
servers
• Routes
servers
• Routes
collector
• Web
portal:
– Looking-‐glass
– Intranet
– Stats
(at
least
SNMP
stats,
but
ideally
sFlow/Ne_low
stats)
– TTS
• Members
mailing-‐list
6