Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Correctness of Flow Migration Across Network Function Instances

Ranjan Patowary, Gautam Barua, Radhika Sukapuram

TL;DR

The paper addresses correct migration of per‑flow state across stateful NF instances and introduces $Weak-O$ as a practical relaxation of stricter order guarantees. It presents a formal system model and a no‑buffering migration algorithm that preserves $Weak-O$, supported by proofs and an implementation on NATs in Mininet showing goodput comparable to non‑migrated paths under common network configurations. Experimental results indicate that reordering can be tolerated thanks to modern TCP variants, making $Weak-O$ feasible in practice. The work also establishes that $Weak-O$ is the strongest no‑buffering, no‑dropping criterion achievable with eventual state synchronization, guiding future NF designs and cross‑NF migration strategies.

Abstract

Network Functions (NFs) improve the safety and efficiency of networks. Flows traversing NFs may need to be migrated to balance load, conserve energy, etc. When NFs are stateful, the information stored on the NF per flow must be migrated before the flows are migrated, to avoid problems of consistency. We examine what it means to correctly migrate flows from a stateful NF instance. We define the property of Weak-O, where only the state information required for packets to be correctly forwarded is migrated first, while the remaining states are eventually migrated. Weak-O can be preserved without buffering or dropping packets, unlike existing algorithms. We propose an algorithm that preserves Weak-O and prove its correctness. Even though this may cause packet re-ordering, we experimentally demonstrate that the goodputs with and without migration are comparable when the old and new paths have the same delays and bandwidths, or when the new path has larger bandwidth or at most 5 times longer delays, thus making this practical, contrary to what was thought before. We also prove that no criterion stronger than Weak-O can be preserved in a flow migration system that requires no buffering or dropping of packets and eventually synchronizes its states.

Correctness of Flow Migration Across Network Function Instances

TL;DR

The paper addresses correct migration of per‑flow state across stateful NF instances and introduces as a practical relaxation of stricter order guarantees. It presents a formal system model and a no‑buffering migration algorithm that preserves , supported by proofs and an implementation on NATs in Mininet showing goodput comparable to non‑migrated paths under common network configurations. Experimental results indicate that reordering can be tolerated thanks to modern TCP variants, making feasible in practice. The work also establishes that is the strongest no‑buffering, no‑dropping criterion achievable with eventual state synchronization, guiding future NF designs and cross‑NF migration strategies.

Abstract

Network Functions (NFs) improve the safety and efficiency of networks. Flows traversing NFs may need to be migrated to balance load, conserve energy, etc. When NFs are stateful, the information stored on the NF per flow must be migrated before the flows are migrated, to avoid problems of consistency. We examine what it means to correctly migrate flows from a stateful NF instance. We define the property of Weak-O, where only the state information required for packets to be correctly forwarded is migrated first, while the remaining states are eventually migrated. Weak-O can be preserved without buffering or dropping packets, unlike existing algorithms. We propose an algorithm that preserves Weak-O and prove its correctness. Even though this may cause packet re-ordering, we experimentally demonstrate that the goodputs with and without migration are comparable when the old and new paths have the same delays and bandwidths, or when the new path has larger bandwidth or at most 5 times longer delays, thus making this practical, contrary to what was thought before. We also prove that no criterion stronger than Weak-O can be preserved in a flow migration system that requires no buffering or dropping of packets and eventually synchronizes its states.
Paper Structure (11 sections, 2 theorems, 7 figures, 2 tables, 1 algorithm)

This paper contains 11 sections, 2 theorems, 7 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Model
  3. Correctness Criteria for flow migration
  4. Preserving Weak-O with N and L
  5. Discussion
  6. Conditions for preserving Weak-O when a flow is migrated:
  7. Implementation of the same NF in two ways
  8. Implementation and Evaluation
  9. Weak-O is the strongest criterion
  10. Related Work
  11. Conclusions

Key Result

Theorem 1

Algorithm alg-migrate preserves Weak-O without buffering or dropping packets.

Figures (7)

  • Figure 1: Properties of flow migration
  • Figure 2: System Model for NF State Migration
  • Figure 3: Sequences of states and substates
  • Figure 4: Flow Migration from one NAT instance to another
  • Figure 5: Exp1: Goodputs with and without migration for various relative delay of old path, for three values of bytes transferred
  • ...and 2 more figures

Theorems & Definitions (25)

  • Definition 2.1: Order preservation (O)
  • Remark
  • Definition 2.2: External-order-preservation (E)
  • Remark
  • Definition 2.3: Loss-freedom(L)
  • Definition 2.4: No-buffering(N)
  • Definition 2.5: Strict Order preservation (SO)
  • Remark
  • Definition 3.1: Correctness criteria for migrating a flow from one NF instance to another
  • Definition 3.2: Partial equivalence of states
  • ...and 15 more