I tried to assist Hazelcast persons redirecting them to my colleagues in order to configure Redis properly, since I was in vacation, but they no longer replied and published this article where, the main point is that they don't know how to configure Redis properly. I think I'll stop forever to collaborate with Hazelcast about benchmarks and let them publish whatever they want: fortunately people chose systems according to other standards, that is, real world testing of properly configured setups and past experiences with other deployment of the same software. They even managed to get stuck in a problem that is not really common when configuring Redis with high load, it is not common to have replication problems because the default output buffer size limit.
How do you suggest one configures Redis to avoid this kind of possible data loss while maintaining throughput? I think when enabling replication, users will assume that recent writes are safe. This bug or whatever you want to call it, seems to allow data to be unsafe as long as the high load persists.