Web Performance Tuning

The Protocols of the Web

Following are descriptions of the most important network protocols for the Web with performance information on each. Lower layers are presented first.

ARP

The Address Resolution Protocol (ARP) is what translates IP addresses to Ethernet’s hardware addresses, also known as Media Access Control (MAC) addresses. A host needing to get an IP packet to a final destination on the same LAN as itself sends out an ARP broadcast asking if anyone knows the MAC address assigned to the desired IP. The machine with the target IP should answer the ARP request.

ARP is cached for a timeout period, and is generally quite efficient. It can be a problem, however, if it is used in place of true routing. Clients should be configured to send packets directly to the local router if no one on the LAN claims them. It may seem clever to have the router send an ARP reply for all addresses (called proxy ARP), rather than just the addresses not on the local subnet, but this puts a big load on the router and network because all packets then have to be examined by the router. See Managing IP Networks with Cisco Routers, by Scott Ballew (O’Reilly & Associates).

PPP

PPP is a link-level protocol for carrying any network-level protocol, unlike SLIP, which carries only IP. If you are using a modem to connect to the Internet directly, chances are you’re using PPP. PPP frames use checksums to flag errors in transmission; a frame with an invalid checksum is retransmitted, slowing perceived speed. ...