End-to-End Network Delay Guarantees for Real-Time Systems using SDN

Presentation of the paper "End-to-End Network Delay Guarantees for Real-Time Systems using SDN", presented at the Real-Time Systems discipline's seminar at DCC/UFMG 2022/02.

Original paper published by RTSS 17: https://ieeexplore.ieee.org/document/8277296

Transcript

1. Sistemas de Tempo Real Lucas Bleme
2. Agenda Problem (motivation) Software Defined Networks Path Finder Algo Performance and effectiveness Limitations Final Evaluation
3. Challenges running RTS on Networks • Lack of Flexibility Changes on flow characteristics might require serious system redesign. • Poor Isolation Non-critical flows could impact on safety-critical system flows delays. Isolation often require hardware redundancy. • Expensive and Complex to manage Existing solutions are mostly proprietary, expensive, requiring dedicated hardware (AFDX and CAN).
4. Proposed Solution An efficient Algorithm to find the best path-layout in any network that meets a given Delay and Bandwidth requirement. Take leverage of SDN global view of the entire network to guarantee end-to-end delays for high-criticality flows (Class I).
5. Software Defined Networks (SDN)
6. Software Defined Networks (SDN) • Network is controlled by software • Vendor agnostic solutions for managing Commodities Switches • Holistically programmable from the controller • Logically centralized intelligence and control https://homepages.dcc.ufmg.br/~mmvieira/cc/papers/minicurso-sdn.pdf
7. Proposed Variation for solving MCP (multi-constrained paths) "Find paths for the flows such that the flow requirements (end-to-end delays and bandwidth) can be guaranteed for the maximum number of critical flows."
8. Proposed Variation for solving MCP (multi-constrained paths) Since MCP is NP-complete, a solution in polynomial-time is proposed: The key idea is relax one constraint at time (delay or bandwidth) and try to obtain a solution.
9. Proposed Variation for solving MCP (multi-constrained paths) Since MCP is NP-complete, a solution in polynomial-time is proposed: The key idea is relax one constraint at time (delay or bandwidth) and try to obtain a solution.
10. Proposed Variation for solving MCP (multi-constrained paths) 1. For each flow starting from the highest priority (delay monotonic), relax the bandwidth requirement. 2. Solves the MCP_HEURISTIC. 3. If a solution exists, assign the corresponding path to the flow 4. If not, relax the delay requirement. 5. If the path is not found after both relaxation steps, return False.
11. Performance of the path finder Algo A topology is schedulable if all the RT flows meet their delay and bandwidth requirements. For each pair (delay and number of flows), 250 topologies are generated and the acceptance ratio is evaluated. 8 x 7 x 250 = 14,000 different topology were tested.
12. Performance of the path finder Algo
13. Effectiveness of the delay guarantees Delay guarantees should be insured even when there are non-critical flows in the network. Enabled 1-3 non-critical flows in the network randomly, during a UDP flow bursting 5 packets per 1ms. Non-critical flows use a separate queue from OVS.
14. Effectiveness of the delay guarantees
15. Limitations • One queue per critical flow, means that better multiplex flow techniques should be used (Open vSwitch limitation). • Enforcing flow priority is not straightforward (delay monotonic), mainly when resources has to be shared. • No admission control for non-schedulable flows.
16. Conclusion RTS can be better managed in a less complex, and more cost effective manner with an SDN approach. Hard Real-Time guarantees (delay and bandwidth) can be insured and isolated on deterministic RTS. The proliferation of commercial-off-the-shelf (COTS) could also benefit critical systems. Innovation in RTS happens when we mix disciplines (Algorithms + Networks).
17. lucasbleme@dcc.ufmg.br https://speakerdeck.com/andreybleme