Overlay Networking Research @NTG
Download
Report
Transcript Overlay Networking Research @NTG
Overlay Networking Research
@NTG
Srinivasan Seetharaman
Fall 2006
Overlay Networks
Overlay networks help overcome functionality
limitations of the Internet by forming a virtual
network on top of the IP network.
Overlay nodes collaborate to offer multiple services:
Multicast (Ex: ESM, Overcast)
Alternate paths (Ex: RON, Detour, X-Bone, Brocade)
Customized forwarding (Ex: I3, Scattercast)
QoS (Ex: OverQoS, SON)
Name resolution, Intelligent routing…
NTG seminar – Fall 2006
Overlay Networks (contd.)
Typical features:
Overlay routing is independent of IP layer routing
(Possibly with functionality overlap)
End-system overlays (Ex: Skype), or
Service overlays (Offer services to actual end-systems)
Each Overlay link is actually an IP route between the two
overlay nodes
Each Overlay path is comprised of one or more Overlay
links, inline with a certain objective
NTG seminar – Fall 2006
Example: Latency-Optimized paths
A
50ms
D
20ms
20ms
Relaying
B
C
Potential for selfish behavior. Impact varies based on the
percentage of overlay traffic
scale of overlay networks
NTG seminar – Fall 2006
What can go wrong?
1.
2.
3.
Overlay routes can violate
native layer polices
- [Srini ICNP06]
Extra load on native links
causing bottleneck issues.
Serious issue when native
layer wants load balance.
- [Srini Ongoing]
A
D
20ms
20ms
Relaying
B
C
Failure detected by both layers and rerouted twice,
with each rerouting disrupting the optimality of the previous.
- [Srini Infocom06]
NTG seminar – Fall 2006
What can go wrong? (contd.)
4.
Sub-optimal performance because
of lack of cooperation. Potentially
a chicken-and-egg problem
- [Srini Hotnets05]
A
D
20ms
20ms
C
5.
Topology (Nodes+Links)
placement is tricky
- [Jinliang Infocom06, Sridhar Ongoing, YongZhu Infocom06]
6.
Overlay routing can clash with TCP congestion control
- [Ruomei Ongoing]
Relaying
B
NTG seminar – Fall 2006
1. BGP Policy Violations
Example violation over Planetlab and its benefit
6.3ms lower latency = 10.48% gain
Transit policy is violated in the above example. Harvard
Univ pays its provider unnecessarily and unavoidably.
NTG seminar – Fall 2006
3. Dual Rerouting
Consider a native link failure in CE
Only overlay link AE is affected.
The native path AE is rerouted over F
(ACE → ACFDE)
G
3
A
I
3
24
OVERLAY
2
E
NATIVE
G
B
I
A
∞
F
C
Cost
Overlay
recovery: 8
Original: 2
Native
Failure
E
D
Native
Rerouting: 2
Overlay
rerouting: 4
Native
Recovery
+
H
Time
Native
Repair
NTG seminar – Fall 2006
4. ESM-Friendly Native Network
A
D
New packet
replication
capability
B
C
Reduce
No extraextra
copies copies
on each on
linksome links
Similar in
spirit to
REUNITE,
Packet
reflection
Latency to reach C is less
… If there were native layer support
NTG seminar – Fall 2006
What can go right?
0.
The services listed earlier, viz. QoS, Multicast, resilience,
intelligent routing, security
Improve performance of TCP using schemes like Overlay-TCP
- [Pradnya Networking05]
Improve services offered at the native layer by deploying
them at the overlay layer (Ex: Name resolution)
- [Sridhar Ongoing]
Creating improved service-based architectures and endsystem driven functionalities
- Anyone interested?
NTG seminar – Fall 2006
References
Jinliang Fan, Mostafa Ammar, "Dynamic Reconfiguration of Service Overlay
Networks," IEEE INFOCOM, 2006. [PDF]
Yong Zhu and M. Ammar, "Algorithms for assigning substrate network.
resources to virtual network components," IEEE INFOCOM, 2006. [PDF]
Pradnya Karbhari, M.Ammar, E. Zegura, "Optimizing End-to-End Throughput
for Data Transfers on an Overlay-TCP Path", IFIP Networking, 2005. [PS]
S. Seetharaman and M. Ammar, "Characterizing and Mitigating Inter-domain
Policy Violations in Overlay Routes," IEEE ICNP, 2006. [PDF]
S. Seetharaman and M. Ammar, "On the Interaction between Dynamic Routing
in the Overlay and Native Layers," IEEE INFOCOM, 2006. [PDF]
S. Seetharaman and M. Ammar, "Overlay-Friendly Native Network: A
Contradiction in Terms?," ACM HOTNETS IV, 2005. [PDF]
NTG seminar – Fall 2006