TL;DR
This paper proposes a "unique nickname" approach to extend TRILL (Transparent Interconnection of Lots of Links) routing to support multilevel network architectures. This method assigns a distinct nickname to every TRILL switch across the entire multilevel campus, simplifying border RBridge operations and potentially improving network management compared to aggregated nickname approaches.
What Problem Does This Paper Address?
The TRILL protocol, designed for efficient layer 2 routing, can be scaled to large networks using a multilevel architecture, similar to how IS-IS routing works. However, implementing multilevel TRILL presents challenges, particularly in managing the unique identifiers (nicknames) assigned to TRILL switches (RBridges) and how traffic is routed across different levels of the hierarchy. Existing approaches, like the "aggregated nickname" method, assign nicknames to entire areas, requiring complex operations at border RBridges to rewrite nickname fields when traffic crosses area boundaries. This paper addresses the need for a simpler, more scalable way to manage nicknames in multilevel TRILL networks.
Methods
The core of this paper is the specification of the "unique nickname" approach. Key aspects include:
- Unique Nickname Assignment: Every TRILL switch (RBridge) in the entire multilevel campus is assigned a globally unique nickname.
- Border RBridge Advertisement: Level 1/Level 2 border RBridges play a crucial role. They advertise to the Level 1 area that certain nicknames are not available for assignment within that area. Simultaneously, they advertise to the Level 2 area which nicknames are already in use by Level 1 areas. This ensures that nicknames are not reused across different areas and that the network knows how to reach nicknames in other areas.
- Simplified Border RBridges: Unlike the aggregated nickname approach, border RBridges in the unique nickname model do not rewrite the nickname field of TRILL data packets when they transition between Level 1 and Level 2. This significantly simplifies their functionality.
- Distribution Tree Calculation: Each RBridge calculates two types of distribution trees:
- Local Distribution Trees: Used for multi-destination traffic confined to a specific area.
- Global Distribution Trees: Used for multi-destination traffic that needs to span across multiple areas.
- Nickname Allocation and Announcement: The paper details how nicknames are allocated and announced. It introduces a new "NickBlockFlags APPsub-TLV" to efficiently advertise the availability of nickname blocks, especially for large numbers of nicknames. This TLV helps border RBridges inform other parts of the network about which nickname ranges are available or in use.
Datasets
Not discussed in the paper. This document focuses on protocol design and specification rather than empirical evaluation on specific datasets.
Key Insights and Ideas
- Global Nickname Uniqueness Simplifies Operations: The most significant insight is that making nicknames globally unique, rather than area-specific, dramatically simplifies the logic required at border RBridges. This reduces complexity and potential points of failure.
- Border RBridges as Information Hubs: Border RBridges are central to the unique nickname approach, acting as conduits for nickname information between different levels of the TRILL hierarchy. Their role in advertising nickname availability is critical for network reachability.
- Two-Tiered Distribution Trees: The concept of distinct local and global distribution trees for multi-destination traffic is a key mechanism for managing traffic scope within a multilevel TRILL network.
- Efficiency in Nickname Announcements: The introduction of the "NickBlockFlags APPsub-TLV" is an important practical improvement, addressing the inefficiency of announcing individual nicknames when large blocks are involved.
Key Results
This paper is a standards-track document specifying a protocol mechanism, rather than presenting empirical results from experiments. The "key results" are the defined protocol extensions and behaviors that enable the unique nickname approach for multilevel TRILL. The primary benefit highlighted is the simplification of border RBridge functionality. The paper also notes that this approach "may be less well suited for very large campuses" compared to the aggregated nickname approach, implying a trade-off between operational simplicity and ultimate scalability.
Limitations and Future Work
- Scalability for Very Large Campuses: The authors acknowledge that the unique nickname approach "may be less scalable and may be less well suited for very large campuses" compared to the aggregated nickname approach. This suggests that for extremely large deployments, the management of globally unique nicknames might become a bottleneck.
- Interoperability with Legacy Devices: The paper mentions that if legacy TRILL switches do not understand the new APPsub-TLVs for tree selection, configurations must ensure that globally used Data Labels are disabled on these devices to prevent incorrect traffic flooding.
- Security Considerations: The paper touches upon security, noting that while multilevel TRILL confines control traffic, the leaking of nickname information between levels can still allow nickname-related events in one area to affect others. It recommends using IS-IS Authentication TLVs to protect sensitive announcements like NickBlockFlags.
- Future Work: While not explicitly detailed as "future work," the document lays the groundwork for implementing and deploying this unique nickname approach. Further research might explore optimizations for extreme scale or more advanced security mechanisms tailored to this specific multilevel TRILL model.
Summary updated