Hello,
I know that everyone is focused on Bristol right now, but Tatiana and Justin both asked that I share the following paper and slides, which were presented at TRANSACT yesterday. They detail our experience in using the features of the specification currently present in g++ to transactionalize memcached.
For what it's worth, I encourage you to focus on the findings/recommendations/issues, and not on performance. Performance *will* improve. The more important issue is that we tried to stick closely to the specification, so that our experiences can be as valuable to SG5 as possible. Note, too, that some issues were not given proper treatment (e.g., exceptions and cancel) since we did not refactor the program from C to C++. Nonetheless, I think it will be a useful (and not particularly long or difficult) read.
- Mike