Forward a PTY device to network

Dependency:
Install latest gevent and ws4py
  pip install gevent --upgrade
  pip install ws4py

Websocket PTY Usage:
0. Start QEMU with isa-serial and chardev as follow
    -chardev pty,id=charserial0 \
    -device isa-serial,chardev=charserial0,id=serial0
  Or add a serial port pty device in virt-manager for the VM

  Boot the VM, when in grub, edit the kernel arguments, append
   console=ttyS0

1. Start ptyforwardWebSocket as the same host running QEMU
    sudo ./ptyforwardWebSocket.py 8888

2. Open wspty/wspty.html in a modern browser

3. In the web page, fill the fields with the following
    Host: 127.0.0.1 or the host ip
    Port: 8888
    PTY dev: /dev/pts/X ,which is the pty device qemu created

  Do not tick Encrypt, click "Connect", press ENTER in the center black screen.


Common PTY forward Usage:
0. Start QEMU with isa-serial and chardev as follow
    -chardev pty,id=charserial0 \
    -device isa-serial,chardev=charserial0,id=serial0
  Or add a serial port pty device in virt-manager for the VM

  Boot the VM, when in grub, edit the kernel arguments, append
   console=ttyS0

1. Start ptyforward as the same host running QEMU
	sudo ./ptyforward 8888

2. Connect to ptyforward
    nc 127.0.0.1 8888
  After connected, input the PTY device path and enter. For example
    /dev/pts/4
  Now the PTY stream is forwarded, type another enter to see the output.

  Or you can run
    ./remoteConsole.sh /dev/pts/4
  to have a raw mode PTY, escape key is ctrl + ].


Problems:
CPU usage is high when the input/output throughput is high. Saving CPU usage is hard because we can not wait for big chunk of data and forward in one batch, instead, we must forward data back and forth instantly. Every time we call os.read from the PTY device, we just get 3 or 5 characters, so when there are a lot of data arriving, os.read is called very frequently, thus the CPU usage is high. To lower the CPU usage, we can add some waits between the os.read calls to accumulate big chunk of data but the user will feel the console is not reacted to the input instantly and may notice the delay. A experimental CPU saving feature is added to the gevent and multiprocess implementation of ptyforward. It is also observed that after we set kernel.sched_min_granularity_ns = 10000000 and kernel.sched_wakeup_granularity_ns = 15000000, which means using coarser schedual time slice, the CPU usage is dropped greately.


Notes:
Contets in wspty sub-directory is copied from project websockify, with some modifications by me to suit into ptyforwardWebSocket.