Existing evaluations of geo-distributed databases still rely almost exclusively on standard limited workloads such as TPC-C and YCSB+T, which reveal little information about the true cost of wide-area coordination. In this paper, we present a configurable benchmarking framework built around the Product-Parts-Supplier (PPS) workload, and use it to evaluate four representative systems that support geo-distributed transactions: Calvin, SLOG, Detock, and Janus. The experiments run across six realistic and insightful scenarios that vary the transactional load, contention, client count, regional bias, network latency, and packet loss.

The results uncover clear design trade-offs between the systems and demonstrate that our new framework is capable of filling several evaluation holes left by the standard workloads. Our framework introduces important features such as dependent transactions that may abort and retry, longer and tunable read/write sets, and fine-grained control over which regions will participate in the commit. However, the PPS workload comes with some limitations, and thus, the framework does not cover every case. As a consequence, for now, it should complement TPC-C and YCSB+T, not replace them.

As modern applications become more global and resource intensive, geo-distributed databases have become critical for fast, reliable data storage. Evaluating the performance of these databases through traditional benchmarks such as TPC-C and YCSB-T is not sufficient to expose all characteristics of the database's performance. A deeper analysis of available benchmarks is needed to determine how to effectively asses geo-distributed databases.

In this paper, we present a modified implementation of the DeathStar hotel benchmark fit for relational databases and use it to evaluate several geo-distributed databases, Detock, SLOG, Janus, and Calvin. We then analyze in detail the unique characteristics of the benchmark and the performance of these databases in the context of the benchmark.

Geo-distributed databases offer the scalability and low latency that contemporary applications demand, but are challenging to implement. It is therefore crucial that they are tested well. Established benchmarks, such as TPC-C and YCSB-T, are limited and do not cover the entire set of workloads that geo-distributed databases are subjected to. In this paper, we discuss the adaptation of DeathStar Movie an existing benchmark for microservice systems for benchmarking these geo-distributed databases. We set up an experiment in which we ran this modified version of the benchmark on four database systems: Detock, SLOG, Calvin, and Janus. The results showed that DeathStar Movie is capable of pushing these systems to their limits. It also showed the different performance characteristics of the systems, stemming from the different ways in which they attempt to overcome the challenges posed by a geo-distributed setting.

Distributed systems are vital for handling large-scale data and rely on geo-distributed databases to ensure low latency and high availability. Traditional benchmarks, such as TPC-C and YCSB-T, are not designed to handle the complexities of geo-distributed environments and do not allow for configuration of multi-home transaction ratios or dynamic data access patterns. To fill this gap, we implement a benchmark based on the MovR workload and assess its performance on the Detock, Janus, SLOG, and Calvin geo-distributed database systems. Key insights revealed through experiments are that network conditions act as a major bottleneck and high concurrency leads to unsustainable latency spikes which severely limits scalability.

In recent years, applications have started using geo-distributed databases, even though their behavior under different workloads remains complex. Therefore, this project analyses how several databases handle transactional workloads using the SmallBank benchmark. We implement and adapt an already existent benchmark, previously used for non-distribuited databases. Furthermore, we use it to evaluate multiple geo-distributed databases highlighting their strengths and weaknesses throughout multiple scenarios designed to stress different parts of the system. We observe that in most scenarios Detock performs slightly better than SLOG, and both outperform Calvin. This aligns with the fact that Detock builds upon SLOG, which itself improves on Calvin. On the other hand, Janus's performance is significantly behind the others due to its communication overhead. However, while the SmallBank benchmark provides an insightful comparison, providing specific advantages compared to previous benchmarks, its adaptation to geo-distributed databases limits its ability to compare scenarios with higher percentages of multi-home and multi-partition transactions.