Roughly speaking: It's a Erlang-like message-parsing platform and the serialization is required to monitor the messages as they flow through the system.There are thousands of processes and messages (objects) sent per second, and along with other meta-data, all are serialized. Its exceptionally efficient until it hits serialization. The messages are schema-less, and user-defined.
If I could speed up the v8:String conversion it would already help tremendously. I'll perform some tests bypassing the UTF16->UTF8 as see what impact that has. Stuart. On Jun 2, 5:18 pm, "bradley.meck" <[email protected]> wrote: > Rather than focusing on scaling serialization, changing architecture might > be more appropriate (using a stream based serializer, removing redundant > data by using references of some kind (even if in userland), etc.). Often > times when we want to share data between things for v8 (I'm assuming out of > process / Isolate) we are going to hit a wall not by the serialization, but > the raw limitation of computing. IE: dictionary construction time. It may > not help you serialize things faster, but a leaner architecture might be a > solution for you, especially if you can isolate out data that can be > processed parallel without relying on each other. Without knowing more about > the problem, I cannot comment beyond that. -- v8-users mailing list [email protected] http://groups.google.com/group/v8-users
