1.
Bandwidth-driven Flow Allocation Policy
for RINA
Michal Koutenský / Vladimír Veselý / Vincenzo Maffione
2020

2.
Motivation
Two related problems:
Guaranteeing bandwidth for flows during their lifetime
→ Reserving part of available bandwidth
Proactive congestion prevention
→ Reject flow allocation requests that could congest the
network

3.
rlite
A light RINA implementation
GPL/LGPL licensed RINA stack for Linux
Logically decoupled into several components
Behavior of components extendable through policies

4.
Bandwidth reservation
Utilizing already available information
Flow requests can specify desired bandwidth
Routing component already keeps a graph of the DIF
We need to extend the graph with capacity information
Keeping our view of the DIF up to date

5.
Bandwidth reservation
Logically centralized, fault-tolerant
Based on Raft consensus algorithm
“Man-in-the-middle” for the flow allocation process
Raft leader intercepts every flow allocation negotiation
Searches the DIF graph to find a viable path
Reserves bandwidth along the path
Raft replication ensures fault-tolerance

6.
Bandwidth reservation
Tied to a specific path
Requires cooperation from the routing component
Raft leader distributes static route updates for each flow
Multi-path routing, better bandwidth utilization
Less predictable paths and connection parameters

7.
Congestion prevention
Utilizing knowledge of bandwidth state
Assume every flow will use all requested bandwidth
No flow may use more than it requested
Reject requests which we cannot satisfy
Centrally managed flow allocation

8.
Congestion prevention
Handling unbounded flows
Two approaches:
● Reject them
● Apply a pre-defined bandwidth limit
Configurable parameter

9.
Congestion prevention
Enforcing rules
The assumption that no flow uses more than requested
is critical
Achieved by rate-limiting every flow

10.
Experiments

11.
Experiments

12.
Questions?