Palantir

1
Palantir

• http//www.csuglab.cornell.edu/Info/People/livshit
  s/Palantir.html
• A window-sharing system for Windows NT
• Max Feingold, Vladimir Livshits, and Yevgeniy
  Rozenfeld

2
General Description

• Palantir is a window sharing system. It allows
  one to view windows opened on a different machine
  on the network.
• We implemented the system as a client/server
  application, in which both the client and the
  server are implemented in the same Windows
  executable. Upon startup, the program places
  itself into the system tray and provides a
  convenient MFC interface for opening client
  connections, sharing server windows, changing
  communication and program settings, etc.

3
Client-Server Organization

• When you run Palantir, it is automatically
  configured to run as both a client and a server.
• All program functionality is accessible from an
  icon that sits in your Windows system tray.
• When you use Palantir as a client, you can
  connect to a server by providing its ip-address.
  The program will query and display the list of
  available widows on that server for you.
• You can select certain application windows
  running on your computer to be shared out
• There is server authentication -- you need to
  supply a valid user name and password for when
  you are trying to connect.

4
Technology

• We are using win32 API bitmap functions to grab
  client area of application windows.
• Bitmaps are created on the server. Since it may
  take a considerable amount of memory and CPU time
  to grab a window bitmap, server is put to sleep
  between subsequent updates. The update frequency
  (refresh late on the picture below) can be
  configured along with many other parameters.
• Bitmaps sent over the network to the the
  client(s) through Windows sockets.

5
Bitmap Processing

• Clearly, it can take a lot of network bandwidth
  to transfer big bitmaps to multiple clients. The
  screenshot saved as a bitmap can take as much as
  1.5Mb with 1024x768 screen resolution. To
  minimize the network load we use the following
  techniques
• Image sampling
• The basic idea is that we dont need to send the
  whole bitmaps -- it is sufficient to only send
  the differences.
• The server maintains the previous bitmap it sent
  to the client in a cache and determines the
  smallest rectangle containing the difference
  difference between the current bitmap and the
  previous one.
• To find this rectangle, we pass over the image
  twice finding the leftmost and rightmost column
  and topmost and bottommost row containing
  changes.
• One of the drawbacks of this approach is that in
  the worst case (i.e., when the two bitmaps we are
  comparing are absolutely identical) we need to
  pass over the image twice, which can be quite
  CPU-intensive for large images.
• To alleviate this problem, we can set up the
  server to look only on each n-th point.
  Effectively, we set up a grid of points we will
  look at. This can considerably reduce the CPU
  load, but this can also affect precision with
  which bitmaps are updated.

6

• We discovered that approach with a grip can
  significantly reduce CPU utilization at the same
  time yielding reasonably precise results with
  many applications.
• However, for applications that require high
  precision (for instance, if you are typing in
  notepad and only every other letter you type gets
  updated on the client) we may end up with a
  pretty distorted image, so we send full window
  updates every so often (the frequency of these
  updates can be configured).
• Another technique we use to reduce network
  utilization is run length compression
• It turns out that typical images we need to
  process have lots of horizontal and vertical
  lines of the same color. Clearly, this is an
  opportunity for run length compression.
• It practice, run length compression performs
  quite well. The compression ratio depends on the
  type of data being transferring for example, a
  2.8MB bitmap of an Explorer window was compressed
  to only 800KB of data (72 compression), while
  much smaller gains were achieved with colorful
  images such as digitized paintings.
• To reduce the CPU overhead of this compression
  technique, it was build directly into the socket
  class. All data that is sent through the socket
  is compressed on the fly.
• Generally, you would want to use some combination
  of these two compression techniques to achieve
  the best CPU/network utilization tradeoffs for
  your situation.
• The program allows you to set various compression
  parameters on a window-per-window basis.

7
Possible Applications

• Sharing of browser windows, images, and documents
  over the network.
• Monitoring student activities.
• Lab administrators monitoring desktop windows on
  many computers from a single workstation.
• Corporations monitoring their employees desktop
  activities.
• Professors demonstrating the use of a particular
  program to students across the network.
• Communication with shared images between graphics
  designers.
• Graphics-based talk applications -- two notepad
  windows open side-by-side.
• Gaining access to resources available on one
  machine, such as a camera installed on a remote
  computer. In one of our experiments, we ran a
  Palantir server on a machine equipped with a
  camera and viewed the output on a computer in a
  different lab.