DC-OTN: Survivability & Cost Efficiency

• DC-OTN is a multilayer optical networking system designed for high-capacity, resilient, and cost-optimized connectivity between datacenters.
• The analysis demonstrates that interlayer backup resource sharing in multilayer survivability can reduce optical layer costs by up to 37% and overall configuration costs by 22%.
• Integrated joint optimization between MPLS and optical layers achieves up to 9% wavelength savings, offering significant CAPEX benefits compared to sequential configuration.

A Datacenter Optical Transport Network (DC-OTN) is a multilayer optical networking system designed to provide high-capacity, resilient, and cost-optimized connectivity for large-scale datacenter interconnects. In DC-OTNs, the integration of an MPLS control layer with a lower optical transport layer enables scalable traffic engineering and robust survivability, supporting diverse traffic granularities, stringent availability requirements, and efficient resource utilization over metropolitan to nationwide domains.

1. Survivability Strategies in MPLS over OTN

Survivability in DC-OTNs is architected via two principal approaches:

• Single Layer (SL) Survivability: Here, every Label Switched Path (LSP) at the MPLS layer is protected by a disjoint backup LSP (pLSP), both logically and physically. This mapping ensures node-link disjointness, but imposes the need to reserve spare capacity for each protected working LSP (wLSP) solely at the MPLS layer. The practical consequence is higher overall transit traffic and a majority of protection overhead realized as additional lightpaths traversing IP routers.
• Multilayer (ML) Survivability: Protection is distributed between the MPLS and optical layers. Specifically, multi-hop LSPs are protected by additional LSPs, whereas single-hop LSPs are protected at the lightpath (optical) level. The optical layer deploys protection lightpaths (pLPs) for physical and cross-connect failures, while the MPLS layer covers router and interface faults. ML survivability supports nuanced spare capacity allocation mechanisms, including:
  • Double Protection: Both working and protection LSPs are mapped onto lightpaths that are themselves protected at the optical layer, incurring high redundancy and resource usage.
  • LSP Spare Unprotected: Only working LSPs are protected optically; protection LSPs are routed over unprotected lightpaths, reducing resource duplication.
  • Interlayer Backup Resource Sharing (BRS): Spare MPLS resources are mapped as pre-emptible (unprotected) traffic in the optical layer without reserving extra capacity, maximizing resource pooling and efficiency.

A fundamental constraint in the survivability ILP formulation is the strict logical and physical path disjointness for each (wLSP, pLSP) pair: $$\forall l \in \mathrm{LSPs}:~ \mathrm{wLSP}(l),~ \mathrm{pLSP}(l)~\text{must be disjoint in both layers}$$ The additional cost for transit traffic traversing intermediate routers is captured as: $$\mathrm{CTT} = c_{p_{\mathrm{IP}}} \cdot (\text{transit traffic})$$

2. Spare Capacity Allocation: Methods and Implications

Spare capacity allocation (SCA) is central to the resilience and efficiency of a DC-OTN:

• SL SCA: All protection is realized at the MPLS layer, compelling allocation for every LSP’s backup. This increases the number of lightpaths and spikes router transit load.
• ML SCA: Methods include:
  • Double protection (maximum resource consumption),
  • LSP spare unprotected (logical-optical decoupling),
  • Interlayer BRS (optimal sharing).

Efficient SCA directly minimizes the number of reserved protection lightpaths and wavelengths. The papers’ analysis quantifies that "interlayer BRS" delivers up to 37% savings in optical layer costs compared to naive double reservation strategies.

3. Cost and Resource Usage Modeling

A comprehensive cost model is employed, combining:

• IP/optical interface cost ($c_{p_{\mathrm{IP}}}$)
• OXC port cost ($c_{p_{\mathrm{OXC}}}$)
• WDM transponder cost ($C_T$)
• Router transit traffic penalty ($\mathrm{CTT}$)

The cost minimization is formalized as: $$\min:~ \mathrm{CTT} \cdot (\text{transit traffic}) + \mathrm{CLP} \cdot [w(i, j) + S(i, j)] + \sum_{(m,n)} c_{\omega} \cdot [w(m,n) + \rho(m,n)]$$ where $w(i, j)$, $S(i, j)$ denote working and spare (protection) lightpaths between nodes, and $w(m,n)$, $\rho(m,n)$ capture working/protection link usage (in wavelengths).

Resource usage pivots on:

• Traffic granularity: Low-bandwidth LSPs require grooming/multi-hop routing, inflating transit costs.
• Physical topology: Sparse networks (low connectivity) force longer, less efficient routing and wavelength consumption.
• Survivability scheme: ML methods, especially with backup sharing, optimize wavelength usage.

4. Comparative Evaluation: Single Layer vs. Multilayer Strategies

Empirical and analytical evaluation indicates:

• Low-bandwidth LSPs: SL survivability is generally superior due to minimized lightpath overhead and reduced cost from unnecessary MPLS protection of single-hop LSPs.
• High-bandwidth LSPs: ML survivability prevails as most flows become single-hop (optimal for optical-layer protection); backup sharing further amortizes protection overhead.

Cost efficiency and resource metrics thus depend on both traffic regime and physical network density.

5. Configuration and Optimization Approaches

Two principal configuration paradigms are considered:

• Sequential (Overlay):
  • MPLS logical layer is provisioned first.
  • Optical layer is configured to realize the logical topology (routing, wavelength assignment).
  • No feedback loop between layers; suboptimal mapping may result.
• Integrated (Peer):
  • Joint (layer-aware) optimization (global ILP) of logical and physical resources.
  • Empirically yields up to 9% wavelength savings over the sequential method.
  • Associated increase in computational complexity for large network instances.

Given that lightpaths often dominate configuration costs (~70%), even single-digit wavelength savings can render significant economic impact.

6. Cost Savings and Scalability for Nationwide Networks

The model incorporates a test network with 12 nodes and population-weighted traffic matrices, representative of metropolitan to nationwide DC-OTN deployment scenarios. Key findings:

• ML SCA methods (especially "interlayer BRS") achieve up to 22% total configuration cost reduction and up to 37% savings for the optical layer.
• Integrated joint optimization brings a further 9% improvement in wavelength usage, albeit with incremental total cost reduction (about 1.4–3%) due to the dominant weight of lightpath infrastructure in CAPEX.

Assumptions such as transparent optical transport with wavelength conversion and resource over-provisioning ensure that observed improvements are primarily cost, not capacity, driven.

7. Practical Recommendations and Design Implications

The findings yield explicit design guidance:

• Employ SL survivability with fine-grained, low-bandwidth traffic, especially in resource-rich, low connectivity DC-OTN topologies.
• Use ML survivability with backup resource sharing for high-bandwidth (single-hop) LSPs to maximize wavelength and cost efficiency.
• Prefer integrated configuration where feasible to optimize wavelength utilization, but be aware of combinatorial problem scale.
• Map spare MPLS capacity to the optical layer using sharing to drive down both configuration and per-wavelength costs.
• Apply the cost and resource usage model iteratively for evolving traffic and connectivity parameters to ensure continued resource and OPEX efficiency at national scales.

The analysis substantiates that judicious, multilayer survivability and spare capacity allocation—augmented by integrated multi-layer optimization—are essential for resource, cost, and wavelength-efficiency in large-scale, survivable Datacenter Optical Transport Networks (0710.4261).