summaryrefslogblamecommitdiffstats
path: root/README
blob: 9074c3ffe0117a3aee68a1d094ad317557c11907 (plain) (tree)
























































































































































                                                                               
INTRODUCTION
============

DNBD (Distributed Network Block Device) is a read-only and caching network
block device and supports following main features:

* replication of servers for robustness
* multicast communication and caching of network traffic for scalability

These characteristics make it suitable especially for use in wireless networks,
e.g. for diskless clients or to share multimedia files in such an environment.
The servers can export a file or block device equipped with a operating system,
movies, music, etc. Several clients can import the block device and access it
like a local hard disk. However, each block transfer over the network can be
cached by all clients: If several users on each client start to watch a movie
within a certain time interval, the movie data has to be transmitted only once
(depending on the cache size). The network is not burdened with unnecessary
traffic. 
DNBD can be used together with cowloop [1] or unionfs [2] in order to get local
write semantics, e.g. for diskless clients. Especially in wireless environments
with limited bandwidth, caching can increase boot-up time enormously.

COMPILATION
===========

DNBD was developed for kernel 2.6.13 and later releases. Kernel 2.4 is not
supported. The kernel sources and common tools (gcc, make, etc.) have to be
installed.

Server and Client
--------------------
Unpacking:
$ tar xzvf dnbd.tar.gz
   
Compiling:
$ cd dnbd; make

USAGE
=====

Server
------
To show available command line parameters, start the server without
arguments:

$ ./server/dnbd-server
dnbd-server, version 0.9.0
Usage: dnbd-server -m <address> -d <device/file> -i <number> 
                  [-t <threads>]

description:
  -m|--mcast     <multicast address>
  -d|--device    <block device or file>
  -i|--id        <unique identification number>
  -t|--threads   <number of threads>

With the following command, the server will be started for the multicast
network with address 239.0.0.1 and export the given file or block device.
Its unique id is 1:

root@server1 $ ./server/dnbd-server -m 239.0.0.1 -d <partition/file> -i 1

To start a server on another computer, the used file or block device must have
the same content and size as on the first server. However, the id has to be
changed:

root@server2 $ ./server/dnbd-server -m 239.0.0.1 -d <partition/file> -i 2

If DNBD is used for wired networks and on multi-processor machines, the
number of threads should be increased to the number of CPUs.

To access the exported file or block device, another computer is used as 
client.

Client
------

The kernel module has to be loaded, before the client application can be used:

root@client1 $ insmod ./kernel/dnbd.ko

There should be an entry in syslog after successful loading. With no command
line arguments the client gives available options:

root@client1 $ ./client/dnbd-client
dnbd-client, version 0.9.0
Usage: dnbd-client -d device -b <address> [-c <file>]
    or dnbd-client -d device -u
    or dnbd-client -d device -c <file>

description:
  -d|--device    <device>
  -b|--bind      <multicast-address>
  -u|--unbind    
  -c|--cache     <file>

We will now import the block device of the server, e.g.:

root@client1 $ ./client/dnbd-client -d /dev/dnbd0 -b 239.0.0.1

The client should tell you that it found a server with id "1". If you exported
a CDROM with a movie, you can watch it on the client over the network, e.g.
with mplayer (usually after mounting).

If someone else wants to watch the movie on a different client, you should
enable caching either during operation

root@client1 $ ./client/dnbd-client -d /dev/dnbd0 -c <cachefile>

or at the beginning

root@client2 $ ./client/dnbd-client -d /dev/dnbd0 -b 239.0.0.1 -c <cachefile>

To create a cache with, e.g. 32M use

root@client1$ dd if=/dev/zero of=/tmp/cachefile bs=1M count=32

Cache statistics are shown with

root@client1$ cat /proc/driver/dnbd/dnbd0

The block device has to be unbound before the module can be unloaded:

root@client1 $ ./client/dnbd-client -d /dev/dnbd0 -u
root@client1 $ rmmod dnbd

FILES
=====

Client
------
./client
   client.c		# client application

Server
------
./server
   net.c		# network routines
   query.c		# server request handling
   filer.c		# file/device I/O
   server.c		# server application (main file)
   
Kernel module
-------------
./kernel
   net.c		# server management
   queue.c		# queue handling for requests
   cache.c		# cache implementation (red-black trees)
   main.c		# module and block device (un)registration, threads


[1] http://www.atconsultancy.nl/cowloop/
[2] http://www.fsl.cs.sunysb.edu/project-unionfs.html