Leaf-Spine Network Fabric Overlay Mapping Cache
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Scalability and performance issues in large, high-performance data centers arise from the need to map tenant endpoint addresses to locator addresses across multiple tunnel end-points, leading to latency, capacity, and cost challenges, as well as inconsistencies in state updates and routing inefficiencies in existing overlay network solutions.
Innovation Solution
A combination of a centralized mapping database and caching mechanism at the tunnel end-point, utilizing a leaf-spine network topology with spine devices hosting a global database and leaf devices maintaining a local cache, enables efficient packet forwarding and minimizes updates during endpoint moves, while separating identity and location to support workload mobility and routing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If mapping state is propagated to all TEPs all the time, then mapping availability is improved, but network latency and bandwidth consumption increase
Solution Approach 1:
The patent segments the centralized mapping database into distributed mapping caches located at each TEP. Each TEP maintains a local copy of mapping state, eliminating the need to propagate entire mapping databases across the network. This segmentation allows TEPs to perform local mapping lookups without contacting a central entity, thereby reducing network latency while maintaining mapping availability through local caching.
Solution Approach 2:
The patent implements preliminary action by pre-populating mapping caches at each TEP with relevant mapping state before it is needed. Through mechanisms such as mapping state push notifications and periodic updates, TEPs receive and store mapping information in advance, allowing them to perform immediate local lookups without waiting for central database queries, thus reducing operational latency.
2Stability of the object's composition
If centralized database is used to hold authoritative mappings, then mapping consistency is improved, but system capacity and cost increase
Solution Approach 1:
The patent segments the centralized database architecture into a distributed cache system where mapping state is replicated across multiple TEPs. Each TEP maintains a local mapping cache that stores copies of mapping state, distributing the storage burden across the network infrastructure. This segmentation eliminates the need for a single large centralized database, reducing system capacity requirements while maintaining consistency through coordinated update mechanisms.
Solution Approach 2:
The patent implements copying by creating and maintaining duplicate copies of mapping state at each TEP's local cache. Instead of requiring all TEPs to query a single centralized database, each TEP holds local copies of the mapping information it needs. This copying approach distributes storage requirements across multiple nodes, reducing the burden on any single system component while maintaining data availability and consistency.
3Reliability
If mapping state is pushed to large number of TEPs, then mapping update reliability is improved, but failure risk and inconsistency increase
Solution Approach 1:
The patent implements feedback mechanisms where TEPs report their mapping cache status and update reception status back to the mapping management system. This feedback allows the system to track which TEPs have successfully received and applied mapping updates, enabling it to identify and re-transmit updates to TEPs that failed to receive them. This feedback loop improves mapping update reliability while maintaining consistency by ensuring all TEPs eventually receive the same updates.
Solution Approach 2:
The patent implements beforehand cushioning by implementing acknowledgment and confirmation mechanisms before considering mapping updates as complete. TEPs must confirm successful receipt and application of mapping updates, and the system maintains rollback capabilities or version tracking to handle cases where updates fail. This cushioning approach prevents inconsistent state by ensuring updates are either fully applied or properly reverted, rather than leaving the system in a partially updated state.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
Disclosed herein are methods of forwarding packets on a network, such as a leaf-spine network having leaf devices and spine devices. The methods may include receiving a packet at an ingress leaf device, and determining based, at least in part, on a header of the packet whether the packet is to be transmitted to a spine device. The methods may further include ascertaining based, at least in part, on a header of the packet whether to perform encapsulation on the packet, encapsulating the packet according to a result of the ascertaining, and then transmitting the packet to a spine device according to a result of the determining. Also disclosed herein are network apparatuses which include a processor and a memory, at least one of the processor or the memory being configured to perform some or all of the foregoing described methods.