A number of different quantities can be measured when examining a web server's performance. This section provides a brief description of each.
The number of requests per second that the web server is able to sustain. Useful as a raw measure of performance. Remember that in measuring this, one can choose either to send a large number of requests down a single connection (using HTTP/1.1 persistent connections) or initiate a new connection for each request. Real world servers typically see between 5 and 20 requests per connection, so if you are using persistent connections, it is sensible to limit the number of requests sent down each connection to around this value.
The number of simultaneous connections the web server can handle without errors is an important metric. Some real world servers have to deal with tens of thousands of simultaneous connections.
The number of bytes transferred per unit time. Usually useful only in conjunction with other metrics (it's easy to achieve high throughput figures when transferring a single large file - less easy to do so when trying to transfer numerous small files down a large number of concurrrent connections).
This metric is used in the SPECweb99 benchmark. The benchmark measures the number of simultaneous connections that the server can support; in order for a connection to be consider 'conformant' it must sustain a throughput of 320,000 bits per second.