It depends how you structure your data. It's possible to have one "last pointer" log for example which is updated synchronously and points at new values in multiple structures which are updated independently/optimistically. This has different timing/properties than using a single mutex.