1. Network Control Jennifer Rexford
Fall 2018 (TTh 1:30-2:50 in Friend 006)
COS 561: Advanced Computer Networks

2. Traditional Division of Labor
Management: policy
Measurement and configuration of switches
Human timescales
Control: events
Among the switches
Distributed state and computation
Data: packets
Local to the switch
Handle each packet
E.g., forward, drop, buffer
Management

3. Example: Shortest-Path Routing
Management: configure the link weights
Control: compute shortest paths on the graph
Data: forward packets to next hop along shortest paths 2 1 3 1 4 u 2 1 5 4 3 6 s

4. Link State: Dijkstra’s Algorithm
Flood the topology information to all nodes
Each node computes shortest paths to other nodes
Initialization
Loop
S = {u}
for all nodes v
if (v is adjacent to u)
D(v) = c(u,v)
else D(v) = ∞
add w with smallest D(w) to S
update D(v) for all adjacent v:
D(v) = min{D(v), D(w) + c(w,v)}
until all nodes are in S
Used in OSPF and IS-IS

5. Link-State Routing Example3 2 1 4 5 3 2 1 4 5 3 2 1 4 5 3 2 1 4 5

6. Link-State Routing Example (cont.)3 2 1 4 5 3 2 1 4 5 3 2 1 4 5 3 2 1 4 5

7. Link State: Shortest-Path Tree
Shortest-path tree from u
Forwarding table at u 3 2 1 4 5 u v w x y z s t
link v
(u,v) w
(u,w) x
(u,w) y
(u,v) z
(u,v) s
(u,w) t
(u,w)

8. Distance Vector: Bellman-Ford Algo
Define distances at each node x
dx(y) = cost of least-cost path from x to y
Update distances based on neighbors
dx(y) = min {c(x,v) + dv(y)} over all neighbors v v 2 y 1 3
du(z) = min{c(u,v) + dv(z),
c(u,w) + dw(z)} 1 x 4 u z 2 1 5 t
Used in RIP and EIGRP w 4 3 s

9. Distance Vector Example2 y 3 2 1 4 5 u v w x y z s t
dv(z)= min{2+dy(z), dx(z)}
= 3 1 3 1 x 4 2 u 1 z 5 t w 4 3 s
dy(z)=1
dx(z)=4

10. Distance Vector Example (Cont.)3 2 1 4 5 u v w x y z s t
dw(z)= min{1+dx(z), ds(z),
2+du(z)}
= 5 3 2 1 4 5 u v w x y z s t
du(z)= min{3+dv(z), dw(z)}
= 6

11. What is the Right Division of Labor?
Management

12. Control Plane as a Distributed System
Two aspects of the control plane
Distributed state (e.g., topology discovery)
Control logic (e.g., shortest path)
Key idea in the ONIX paper
Leverage existing distributed systems techniques
Create a state discovery and management layer
Support different consistency & durability properties

13. Hop-by-Hop Utilization-aware Load-balancing Architecture
Naga Katta, Mukesh Hira, Changhoon Kim, Anirudh Sivaraman, and Jennifer Rexford

14. HULA Multipath Load BalancingS2
ToR 10
Data S3
ToR 1 S1 S4
Load balancing entirely in the data plane
Collect real-time, path-level performance statistics
Group packets into “flowlets” based on time & headers
Direct each new flowlet over the current best path

15. Flowlet Routing Using the flowlet table
Dest ToR
Timestamp
Next-Hop
ToR 10 1 S2
ToR 0 17 S4 …
h(flowid) 1 … S3 S1
Data
Data S4
Using the flowlet table
Update the next hop if enough time has elapsed
Update the timestamp to the current time
Forward the packet to the chosen next hop

16. Path Performance Statistics
Best-hop table
Best Next-Hop
Path Utilization S3
50% S4
10% …
Dest ToR 1 …
Probe
Probe S3 S1
Data
Data S4
Using the best-hop table
Update the best next-hop upon new probes
Assign a new flowlet to the best next-hop

17. Putting it all Together Using P4
data packet
Best Next-Hop
Path Utilization S3
50% S4
10% …
Dest ToR 1 …
current best next-hop S3
Dest ToR
Timestamp
Next-Hop
ToR 10 1 S2
ToR 0 17 S4 …
Update next-hop (if enough time elapsed) and time
h(flowid) 1 …
chosen next-hop