Re: User Space question

... performance and its myths.
Reading about DataQ performance compared to sockets or DTAQ versus database files, or database with or without journaling or RLA versus SQL you could read very diffrent recommendations, but no reference information or real benchmarks. World isn't that simple as many people are thinking. The performance bottleneck for Dataqs is on Q read by many clients, the switch over has to be synchronized. I don't know, what's faster dataq synchronisation or database synchronisation, in my release, in the current release or in an upcoming release. But I know very well what's fast enough for me. We've used the central key file aproach in a massive parallel load process in a BI environment and with some blocking (fetching 20 keys at once, giving it out one by one in this job) we've had unlimited scalability. We adjusted the parallelism until all ressources where used, to get maximal throughput and we reached about a million complex transactions equivalent to about 30 Million inserts and updates (most of them inserts) per hour. BTW: all files where journaled, all jobs used commitment controll.