here are some case studies which have "massive bursts of writes".
MIT Lincoln Lab: Real-time Sensor Data Warehouse Architecture Using MySQL
http://conferences.oreillynet.com/presentations/mysql05/nikom_jacob.pdf
http://www.mysqluc.com/cs/mysqluc2006/view/e_sess/8135
The RTS Operations Coordination Center (ROCC) manages the streams of data from multiple sensors, processes the acquired data, and provides operator displays to track and predict the path of space objects. To perform these tasks ROCC needs to implement real-time data storage and distribution functionality.
Real-time storage of high rate data streams coming from multiple ROCC sensors is a major database management system (DBMS) challenge. The DBMS cannot lose any information, but at the same time it cannot be a bottleneck for the whole system. This ability can be evaluated by measuring the database server throughput.
Throughput benchmark was implemented more than two years ago to compare MySQL server performance on different operating systems and hardware. We will discuss the benchmark and the results which show excellent throughput performance of the MySQL server.
The real-time data distribution solution comes from the possibility to create a DBMS-based messaging system with very low latencies. Such a concept could bring together data processing and distribution, unify API to access the latest and oldest information, and centralize data management. The DBMS needs to provide a very fast response to data requests while allowing message filtering on the server side. This ability could be evaluated by measuring the delay between storage and retrieval times or latency performance.
Recently, the latency benchmark was implemented to evaluate MySQL server performance in a real-time messaging system with many writing and reading clients. We will discuss the benchmark design, execution, and the results which indicate good responsiveness of the MySQL server.