Abstract
Implementing geographical redundancy in distributed systems aims to mitigate latencyissues for a global audience by strategically placing data and services across multiple geographical regions. This approach not only enhances user experience by reducing response
times, but also improves the availability and reliability of an application. However, designing and managing geographically distributed architectures presents significant complexities. Data synchronization between components spread across multiple locations is
a major consideration, as maintaining consistency becomes increasingly difficult due to
the unreliable nature and higher latency of wide-area networks.
This master’s thesis discusses the setup and challenges of implementing geographical
redundancy in DealerHUB, a large software system used by motorcycle dealers worldwide for retail and servicing operations. A prototype similar to DealerHUB is developed
and deployed to Azure. Two commonly used architectures for building geographically
distributed applications are explored. The first approach features the Geode pattern,
which allows services to be deployed to multiple geographical regions, with all regions
connected to a common data backplane supporting multi-region writes. This enables
the system to stay operational even if one region becomes temporarily unavailable.
The second approach involves distributing read-only replicas across multiple regions,
enhancing read performance by serving requests from nearby replicas. However, write
requests must still be routed to the primary region, due to the lack of multi-region write
capabilities.
The study evaluates the performance and availability of each architecture through a
series of experiments and performance tests. The results indicate that while read-only
replicas enhance the global read performance, write operations still encounter latency
issues. A robust multi-region write strategy, possibly involving a migration to Azure
Cosmos DB for NoSQL, is recommended for large-scale applications like DealerHUB to
ensure efficient data synchronization and consistency across multiple regions.
| Date of Award | 2024 |
|---|---|
| Original language | English (American) |
| Supervisor | Johann Heinzelreiter (Supervisor) & Peter Ginter (Supervisor) |