Sorry Dustin, I don't recall the exact incompatibility, just that it didn't work for what we were using it for. It may well have been the node weighting/distribution differences, or even the behavior when you lose a node. I can't recall. Here's where we originally talked about it: http://lists.tangent.org/pipermail/libmemcached/2008-September/000435.html
On Sun, Nov 16, 2008 at 4:53 PM, Dustin <[EMAIL PROTECTED]> wrote: > > > On Nov 16, 4:42 pm, "Garth Patil" <[EMAIL PROTECTED]> wrote: > >> that the consistent hashing implementation in libmemcached was >> incompatible with spymemcached (Java), and both were incompatible with >> the original ketama implementation provided by last.fm >> (http://www.last.fm/user/RJ/journal/2007/04/10/rz_libketama_-a_consist...). > > Can you give me an example of what you mean by incompatibility? I > generated a set of results with libketama and made a unit test out of > them guaranteeing that I get the same results for the same input. I > can't produce a .jar unless that's true. > > The only place where there's any incompatibility I know about is > that I don't support node weighting. Someone has put that together in > a branch on github, but I haven't put the effort into cleaning it up > enough to incorporate it into a release. Is this what you're talking > about?
