In the case of a P2P application, applications don’t always run on the server.

• Arbitrary end systems directly communicate with one another
• Peers request service from other peers, which provide service to other peers
  • Self scalable ⇒ new peers bring in new service capacity and service demands
• Peers are intermittently connected and change IP addresses
• e.g. BitTorrent, streaming, VoiP

File Distribution Time

In a client-server application, server must sequentially upload $N$ copies of a given file.

• $F/u_s$ for one copy, $NF/u_s$ for $N$ copies

Each client must also download the file, bottlenecked by slowest client download

• Slowest client download time: $F/d_{min}$, where $d_{min}$ = min client download time

Time to distribute file $F$ to $N$ clients with client-server approach is thus given by:

$$ D_{c-s} \geq \max\{NF/u_s, F/d_{min}\} $$

• Note this increases linearly in $N$

In P2P system, main difference is that clients as aggregate can combine sharing capacity:

• Time to send single copy from server is $F/u_s$
• Slowest client download time is $F/d_{min}$
• Clients as aggregate must download $NF$ bits, but max upload rate is (limiting max download rate) is $u_s + \sum_{u_i}$