Proposed: for using workloads to emulate the usage patterns

Context

V-Sekai is a social virtual reality platform. Users interact with each other through avatars in different environments. The main operations include high read operations, moderate to high write operations, and latency sensitivity.

Problem Statement

To ensure the best performance and user experience, we need to understand how the database system might perform under conditions similar to those experienced by V-Sekai. This includes balancing between high read operations for content delivery and write operations for dynamic user interactions, all while maintaining low latency.

Proposed Solution

We propose to use the Yahoo! Cloud Serving Benchmark (YCSB) workload to emulate the usage patterns of V-Sekai. This would involve a combination of workloads:

Workload A

50% read and 50% update operations.
Simulates user interactions within V-Sekai, such as position updates, gestures, or environment interactions.

Workload B

95% read and 5% update operations.
Represents the frequent fetching of environment and avatar details.

Customized Workload

Adjusting the read and write ratios to reflect the real-time, interactive nature of V-Sekai.
Emphasizing low latency in the benchmarking parameters.

1. Set up YCSB with Workload A and B
2. Customize the read/write ratio based on V-Sekai's usage patterns
3. Emphasize low latency in the benchmarking parameters
4. Run the benchmark and analyze the results

Benefits

Provides insights into how a database system might perform under V-Sekai-like conditions.
Helps optimize the balance between read and write operations.
Ensures low latency for a seamless user experience.

Downsides

Requires careful tuning and understanding of YCSB workloads.
The benchmark results might not perfectly reflect real-world usage due to the synthetic nature of benchmarks.

The Road Not Taken

An alternative would be to use pre-defined YCSB workloads without customization. However, this might not accurately reflect V-Sekai’s specific usage patterns.

The Infrequent Use Case

In scenarios where the usage patterns are not consistent or predictable, customizing the YCSB workload might not provide accurate insights.

In Core and Done by Us

The setup and customization of the YCSB benchmark will be done by the V-Sekai development team with support from the community stakeholders.

Status

Status: Rejected

Decision Makers

V-Sekai development team and community stakeholders.