Upload of huge file using a web application

The environment for the given objective is not currenly available, hence, I'm not able to try out things and have to rely on the analysis only !

My objective can be broken into the following distinct steps :

Uploading huge files(upto 100GB) using a dumb 'Upload File' page - there is no escape from this as the users want a (dumb)front-end and are not willing to ftp the file etc.
The web application which provides the above front end will be hosted on a low-end machine - 2GB RAM and 40GB HDD and this web application WILL NOT STORE any part of the huge file on the local machine but must 'quickly' write it to a high-end remote Linux machine

For each step, I'm highlighting my approach,concerns and queries :

I referred this thread which confused me as I was planning to create a dumb web application using Spring MVC with an upload page - do I need to go into the HTML5 etc. or a simple web application will suffice?

Given the 2GB RAM, the web application will get less than 1GB of it. I'm afraid that an 'OutOfMemoryError' is probable if the code is not written strictly - I have to ensure that from the stream, a small chunk, say 10MB must be read at a time and written to the remote Linux machine's file. Assuming that I am in the Controller Servlet's doPost(...), I did some reading about how to proceed and got confused :

      /**
         * @see HttpServlet#doPost(HttpServletRequest request, HttpServletResponse
         *      response)
         */
        protected void doPost(HttpServletRequest request,
                HttpServletResponse response) throws ServletException, IOException {
            // TODO Auto-generated method stub

            InputStream fis = request.getInputStream();
            int i = 0;

            /* Approach - 1 : Plain old byte-by-byte method */
            Socket socket = new Socket("192.168.90.20", 22);
            OutputStream remoteOpStream = socket.getOutputStream();

            while ((i = fis.read()) != -1) {
                remoteOpStream.write(i);
            }

            /* clean-up */

            /* Approach - 2 : NIO */
            ByteBuffer byteBuff = ByteBuffer.allocate(10000);/* read 10MB of data */

            ReadableByteChannel rdbyc = Channels.newChannel(request
                    .getInputStream());

            File remoteFile = new File("192.168.90.20/Remote_Linux_Folder");/*
                                                                             * Dunno
                                                                             * how
                                                                             * to
                                                                             * create
                                                                             * a
                                                                             * File
                                                                             * on a
                                                                             * remote
                                                                             * Linux
                                                                             * machine
                                                                             */
            FileOutputStream remoteFos = new FileOutputStream(remoteFile);
            FileChannel writableChannel = remoteFos.getChannel();

            while (true/* dunno how to loop till all the data is read! */) {
                rdbyc.read(byteBuff);
                writableChannel.write(byteBuff);
            }

            /* clean-up */

        }

I need some way wherein the data storage on the local machine is minimal - the code simply reads n bytes from the input stream and writes the same to a remote machine

I believe NIO is the way to go but I'm not able to establish as to how I must proceed - please guide about the same.

I would implement a FixedSizeQueue and popAll() the data out of the QueueStream to another computer. Probably have it double buffered, just to provide the cushioning for network/bandwidth problems.