Transcript Chervet Benjamin

Floodless in SEATTLE : A Scalable
Ethernet ArchiTecTure for Large
Enterprises.
Changhoon Kim, Matthew Caesar and Jenifer Rexford.
Princeton University
2009, 3rd March
Presented by Chervet Benjamin
Animation on page 9, 14, 15 and 16 come from Changhoon Kim’s presentation.
1
Motivation
Networks are getting larger

2

Up to 25 000 hosts and 1 000 switchs for large
enterprises Networks.

Larger and larger datacenter at Google, Yahoo or
Microsoft for the largest ones.

Everyone in a University has access to the Internet.
Motivation (2)
Networks are getting larger
Hard to keep efficient, recquires networks administrators
and operators.
More money is spent on maintenance than improving the
Network with new materials.
Need to find out where the problems is and how to solve
it.
3
Outline

Find where the problem is : Current Networks




Solve the problem : Seattle architecture



Objectives
Solutions offered
Evaluate the solutions :



Ethernet Network
Hybrid (Ethernet/IP) Network
Comparison Ethernet/IP
Simulation results
Experimental results
Conclusion


4
Benefits
Lessons Learned
Current Networks
Ethernet Network

-Heavily rely on broadcast to know where a host is
(ARP, DHCP)
- large use of bandwidth,
- privacy concerns
+ learn automatically the address,
+ ensure mobility of a machine
-Flat addressing : Switch maintains a table
associating all the Mac Address of the Network
with its output ports.
-can lead to very large table
-Build a spanning tree
+ No loops in the Network
- links unused
- uneven loads
5
Actual Networks
Ethernet Network

6
Actual Networks
Hybrid IP/Ethernet Networks

LAN
-The broadcast stay in a subnet.
+ No flood on the entire Network.
- Manual configuration needed.
- No mobility possible.
LAN
IP
Network
LAN
7
Machine sharing the same subnets. The
subnet are interconnect by IP Networks.
-Routing done though OSPF
+ Efficient use of links
+ No loops
Actual Networks
Hybrid IP/Ethernet Network

8
Comparaisons
Architectures
Features
Ease of configuration
Optimality in addressing
Host mobility
Path efficiency
Load distribution
Convergence speed
Tolerance to loop
9
Ethernet
Bridging
IP
Routing
SEATTLE





















Comparaisons

SEATTLE :

Get the best of Ethernet and IP.

Configuration free (like Ethernet) and Scalable (as IP).
10
Seattle architecture

Objectives

Avoid flooding



Limits broadcasting


Directory service
Keep forwarding tables small


11
Send a message only to the receiver
Never broadcast unicast traffic
Host location should be kept only where it is needed
Distribution of the data.
Seattle architecture

Objectives (2)

Path efficient.



Backward compatible


12
Need a routing protocol
Switch can learn the topology
Compatible with IP and Ethernet Network
Does not modify end-hosts view
Seattle architecture

Solution offered :

Network layer one hop DHT
All the Switch implement a hash function F, associating
<key, value> pair.
- For every key F returns the same value.
- F returns one of the switch from a key.
- Each Switch must know about all the other living switch.
Cache system is used :the switch cache some frequently used
information
Smart cache update implemented : retrieves the information by
surveying the traffic.
13
How does it works ?
x
Deliver to x
Host discovery
or registration
C
y
Traffic to x
A
Hash
(F(x) = B)
Tunnel to
egress node, A
Entire enterprise
(A large single IP subnet)
Switches
End-hosts
Control flow
Data flow
14
Optimized forwarding
directly from D to A
Tunnel to
relay switch, B
D
LS core
Notifying
<x, A> to D
B
Store
<x, A> at B
Hash
(F(x) = B)
E
Response to host mobility
Old Dst
x
< x, G >
< x, A >
when shortest-path
forwarding is used
A
D
< x, A >
< x, G >
Relay (for x)
New Dst
G
< x, G >
15
B
< x, A >
< x, G >
y
Src
SEATTLE Architecture

Handling ARP requests
4. Broadcast
ARP req
for a
Owner of
(IPa ,maca)
sb
a
1. Host
discovery
sa
2. Hashing
6. Unicast
ARP req
to ra
F(IPa) = ra
5. Hashing
F(IPa) = ra
7. Unicast ARP reply
(IPa , maca , sa)
to ingress
Switch
End-host
Control msgs
ARP msgs
16
b
ra
3. Storing
(IPa ,maca , sa)
Evaluation of the solutions

We want to know if the new solutions performs better
than current networks ?

Simulation using a packet simulator

4 different topologies




17
Campus network (517 routers and switches)
AP-small (87 routers)
AP-large (315 routers)
DC (Data center network)
Stretch: Path Optimality
Stretch = Current path length / Shortest path length
18
Control messages
19
Size of the tables
20
Experimentation on Emulab

Emulab: similar to planetLab: set of PC around the
world.

Useful to test real world networks with some practical
values for latency and bandwidth

In this experimentation: 10 PC Free BSD Nodes
21
Number of messages exchanged
22
Conclusions:

Benefits :




Reduce the flow of data exchanged by order of magnitude
Fast and efficient reaction to changes
Reliability and capacity grows with the size of Network
Less man-made configuration needed
Plug and Playable Networking ensuring efficiency and
scalability.
23
Conclusions

Lessons learned

A new protocols has been created by combining different
technologies from different background.



24
DHT Based routing used first in P2P technologies.
Link state routing
Caching