[Date Prev][Date Next][Date Index]
[thought] Reliable-Multicast
Hello Prof.
I have penned down a few thoughts.. Please let me know if you have any
pointers..
Regards
Shashank
Background:
A growing number of applications rely on basic multicasting support, which
in case of TCP/IP networks is provided by IP (only Class D IP addresses are
meant for multicasting). For e.g. To send a multicast to a subset of hosts
on the internet, the source host needs to send data to a Class D IP address.
All receiver hosts, which intend to listen to this multicast need to
explicitly join this session by sending a JOIN() message to a
multicast-capable router (all current routers support multicast).
Multicasting of UDP streams like audio, or video over the internet is quite
common and you can get one of those multicasts by opening a real player and
tuning to a particular station. This type of multicast is referred to as
unreliable multicast, as loss of some packets can be tolerated by the
receivers..There is another class of applications which cannot tolerate such
packet losses (for e.g. software distribution, Stock updates etc) and
require some kind of reliability in the multicast. In other words they
require the transport layer protocol to be TCP-like rather than UDP.
Problems with reliable multicast:
Some problems associatiated with reliable multicasting using TCP as the
transport layer protocol are:
1. ACK implosion: Every receiver will need to send a ACK to the sender,
which causes a huge traffic of ACK's. A number of proposals can be found in
literature which try to overcome this.
2. The sender can send packets only at the pace of the slowest receiver.
This problem is exacerbated in wireless mobile networks, where the wireless
links can loose packets and the hosts can move.. There are a few papers that
can be found on IEEE that try to deal with this situation, but after reading
a few of them, I felt that they were very bulky (Still need to read a few
others). Theses papers also deal with cases where every node is a potential
source..
Our Problem Space:
We need a lightweight protocol that can effeciently acheive reliable
multicast, for a situation where the mobile hosts can only act as multicast
recievers.. (This is just a special case, and further genralizations are
possible)..
I feel that there are a lot of applications where the mobile nodes are just
passive listeners.. Some of them are listed below.
1. In the commercial sector, updates to stock prices,
advertisements,anouncements etc. usually originate from a wired host and the
mobile host are only receivers to these updates.
2. In the military sector, the mobile nodes are only receivers to updates
regading enemy positions.
I have thought of something, but I am also open to any brainstorming
sessions for such a protocol..