Yusuf Goolamabbas wrote: > > Maybe there is an opening for "mtafs" :) which would have the semantics > which the MTA authors want and reasonable features which are useful to > others [journalling,speed,io-clustering,etc] Queston is: what do the MTA authors want? We occasionally see chunks of fuzzily-informed mud come flying over the wall from MTA lists, but I've yet to see a neat description of the requirements. I'd very much like to see such a description, and to have such a discussion, because we should have a good fit here. > The question is what precise semantics of ffs does ext3 and > ffs/softupdate not provide which makes running a MTA on any recent OS > similar to playing russian roulette with mail One man's `semantics' is another man's `side-effect' :) We've never had an objective of retaining side-effect compatibility with ffs. In ext3, if you want something sync'ed, you sync it. Now we do have a useful side-effect, and that is that if you sync something, you've just synced everything which preceded it. That's not a thing which is going to change in the 2.4 kernel timeframe, and I doubt that it'll ever change, given that it's the most efficient way of handling any pending write data. Still. Please: what are the requirements for MTAs, and for postfix?
