Pipe is similar to IPC and remoting is generally used as RPC. So Pipe will always be faster due to simple memcpy() in SMP, compared to serialize-send-receive-deserialize steps.
Hope you got the point. On Jun 7, 9:31 pm, Jayhawk <[email protected]> wrote: > I have existing code which takes commands over a pipe and returns > large amounts of data. I didliked the design which was not as compact > as I would like, so I rewrote it to use remoting. The resulting data > transfer speed is now 12 times slower than using the pipe. > > I do use a binary tcp channel. I make a singleton object available > which lets the remote client open a binary datafile (it is a hd video) > and request sequential blocks of data (frames) from the server. This > works well, but very slowly compared to just streaming the binary > data, using binary stream writer, over a pipe. > Note that currently both client and server runs on the same computer. > > The question is now if this is to be expected. Should one generally > avoid using remoting for datatransfer intensive tasks, or might I > simply be doing something silly somewhere?
