Toward a Theory of Input Acceptance for Transactional Memories

摘要

We upper-bounded the input acceptance of well-known TM designs and we proposed a new TM design with a higher acceptance. Our conclusion is that accepting various workloads requires complex TM mechanisms to test the input and to possibly reschedule it before outputting a consistent history. We expect this result to encourage further research on the best tradeoff between design simplicity and high input acceptance.