Beruflich Dokumente
Kultur Dokumente
Rou)ng
and
Infrastructure
Direc)ons
Gaurab
Raj
Upadhaya
gaurab@llnw.com
About
Myself
• Network
Architect
at
Limelight
Networks
• Previously
(2002-‐2010),
Sr.
Internet
Analyst
at
Packet
Clearing
House
• Focus
on
Backbone
Opera)ons,
Internet
Exchange
Points,
Research
on
rou)ng
and
peering
rela)onships
Future
Rou)ng
• Different
Perspec)ve
– Technical
• IPv4
/
IPv6,
LSNs
• Large
FIB
opera)ons
•
Mobility
– Commercial
/
Business
driven
• Transit
• Peering
• Other
deals
Technical
• Are
rou)ng
protocols
as
resilient
in
IPv6
as
they
have
been
in
v4.
– Not
the
standards,
but
the
code
and
implementa)on
– Best
Prac)ces
(or
lack
thereof)
– Are
your
router
memory
and
processing
power
enough
for
the
larger
address
pool
– Different
way
of
thinking
needed
..
LSN
/
CGNs
• The
deployment
of
v6
will
bring
in
usual
challenges
• What
should
be
the
‘transi)ons’
mechanism
• Flavours
of
NAT
–
444,
646,
464
…
and
others
• All
of
them
will
need
to
do
n
to
n
transla)ons
at
large
scale,
as
there
are
not
enough
v4
address
to
go
around.
• Thus
the
need
to
consider
what
is
referred
to
as
‘Carrier
Grade
NAT’
or
in
neutral
terms
‘Large
Scale
NAT’.
• This
is
primarily
in
the
edge
and
access
networks
• Deploying
v6
is
possibly
the
only
way
forward
Large
FIB
• How’s
the
address
delega)on
to
customers
going
to
work
?
– Ques)ons
s)ll
con)nue
about
/64
vs.
/48
and
any
unit
in
between.
– How
are
you
going
to
carry
them
in
your
IGP
?
– Internal
aggrega)on
or
/64s
in
FIBs
?
• Newer
Hardware
Design
Model
?
– The
aggregate
of
external
interfaces
are
same
as
internal
b/w
of
the
hardware
Experience
and
BCP
• Big
challenges
will
be
in
the
form
of
not
having
enough
experience
with
the
rapid
changes
• Virtualiza)on
of
services
bring
in
security
challenges
and
forensics
• User
demand
and
flashmob
can’t
be
expected,
and
it’s
much
more
tricky
to
shie
large
amount
of
streaming
data
instantly.
• Newer
applica)ons
are
designed
for
high
bandwidth,
will
create
more
issues
in
the
‘EDGE’
countries.
– Not
just
the
Nepal,
Laos,
Fiji
of
the
worlds,
but
also
in
places
like
New
Zealand,
Thailand,
Veitnam
Mobility
• Mobility
brings
a
big
challenges
• Users
are
always
ON,
and
operators
not
only
need
to
op)mize
their
mobility
rou)ng,
but
also
manage
the
spectrum
use
• Will
providers
go
back
to
walled
gardens
?
– Users
think
of
mobiles
as
an
extension
of
their
desktops/laptops
• Facebook
enabled
SIM
–
is
that
the
future
?
Commercial
/
Business
• This
is
what
will
drive
the
Internet
in
the
next
genera)on
• Network
infrastructure
and
resul)ng
rou)ng
policies
are
mostly
result
of
commercial
limita)ons
• Access
network
providers
have
different
needs
then
other
users
like
banking,
gaming,
soeware
developments
etc.
What
is
important
to
whom
• Access
and
ISPs
– Much
traffic
is
now
geared
towards
the
‘hyper
giants’
• Peer
off
the
content,
and
deploy
caches
– Gamers
care
about
latency
• Try
to
get
closer
to
the
content.
– Bankers
and
financial
industry
wants
lower
latency
• Buy
capacity
on
direct
runs
to
financial
hubs
like
Hong
Kong.
• Research
says
<3ms
and
you
have
a
customer.
Peering
• Peering
is
when
you
don’t
pay.
• But
you
s)ll
need
to
pay
for
the
backhaul
• Unless
the
provider
is
part
of
the
consor)um
laying
undersea
cables,
backhaul
can
be
cost
prohibi)ve
• While
networks
in
the
‘hub’
regions
of
the
Internet
have
been
peering
for
long
)me,
the
‘fringe’
countries
are
finally
gelng
there.
– How
does
the
pay
off
between
free
peering
combined
with
backhaul
cost
with
lower
latency
compare
with
buying
IP
transit.
Transit
• Transit
is
a
slow
market
and
in
a
downward
slope
• Backhaul
is
where
the
large
providers
are
focusing
anyway
• Transit
cost
in
major
markets
are
constantly
going
down
• Ask
some
of
the
people
in
Peering
Forum.
Bandwidth
:
Backbone
Vs.
Edge
• How
much
bandwidth
do
you
build
into
the
core,
when
your
users
are
connected
at
1G.
• Tradi)onally,
ISPs
have
oversubscribed
the
edge
and
varia)ons
in
usage
to
make
money
•
But
with
‘always-‐on’
services
and
high
mobility
use,
their
tradi)onal
models
will
be
under
pressure
• Do
you
provision
more
or
start
believing
in
QoS/TE.
?
Content
Delivery
Rou)ng
• CDNs
all
have
their
rou)ng
priori)es
based
on
their
customer
needs
– Customers
needs
may
not
be
same
as
end
users’
ISP.
• Some
care
about
latency
,
some
don’t.
• Some
have
larger
footprint,
some
don’t.
• The
larger
the
traffic
they
push,
they
start
becoming
more
network
heavy
Latency
Vs.
cheap
path
• Network
Providers
have
a
hard
)me
with
diverse
demands
from
their
users
– High
paying
/
higher
margin
customers
like
financial
industry
wants
lowest
latency
– Low
paying
/
lower
margin
volume
residen)al
customers
don’t
really
care,
but
need
large
amounts
of
bandwidth
• Where
is
the
balance
?
Anycast
Services
• DNS
and
cri)cal
infrastructure
are
already
heavily
anycasted
– UDP
anycast
is
well
known
– TCP
anycast
is
proven
to
be
resilient
• But
then
it’s
a
problem
with
long
lived
tcp
sessions.
• CDNs
to
a
large
extent
depend
on
DNS
services
to
direct
users
to
nearest
POP,
and
combined
with
more
anycast,
this
leaves
a
lot
on
the
soeware
side.
Other
Issues
for
operators
• Lawful
Intercepts
– Should
route
through
black
boxes
or
mirror
traffic
?
• Scaling
– What
to
build
for
?
• Variety
of
planorms
to
deal
with
– DSL,
broadband,
Mobile
• B/W
crunch
Where
will
we
go
?
• Think
of
scaling
and
scaling
and
scaling
– From
operators
perspec)ve,
they
don’t
know
where
to
build
to
scale
• Innova)on
is
coming
from
various
places
– Operators
can’t
keep
up
• Cri)cal
services
are
increasing
– Will
it
mean
more
government
monitoring?
• Are
ISPs
becoming
‘commodity’
services
?
Is
metered
services
the
answer?
• Or
what
is
?
Thanks