Re: about the (bad) twin problem solution

In article <9565@emory.mathcs.emory.edu> mm@ssf.pt (Manuel Monteiro) writes: [I edited the included file] >Hi, > >We are in big trouble caused by the informix solution to the twin >problem. > >There are lots of processes that insert rows into that table. Since >the key values always grow (serial column), there's no reason to the >twin problem to become a problem to us. > >What really happens is that when some process inserts a value, say X, >in an indexed column with immediately lower value Y, no other process >can insert a row with a value on this column between Y and X, until >the first one finishes its transaction. >This represents a big bottle-neck to us, since we have very long >transactions. > >Since there's no reason to deal with non-unique indexes as with unique >ones, in the resolution of the twin problem, I think this is a known >bug since Informix-Turbo and I'd expect it to be solved at least on >OnLine 5.0, but it is not. > >So, if anyone knows some workaround for this 'twin' problem, please let >me know. > >Informix, when will this bug be fixed? > >Thanks. > >| Manuel Figueiredo Monteiro The problem of concurrency with INSERTs due to adjacent key locking is Bug 13674. It will be greatly improved in all OnLine 5.01 ports, at least. Craig