Persistent HTTP: without and with pipelining

A web page with N objects on the same server as the HTML base file, downloaded over a single persistent TCP connection. Change the parameters and compare both delay diagrams on the same time scale.

Link

Data rate R

d_prop

RTT = 2 · d_prop. Requests and TCP control segments are so small that their transmission time is neglected.

Web page

The browser first downloads the HTML base file; only when it has received it does it know which N objects to request.

HTML size

Object sizes

Objects N

Size L

Sizes

One size per object, separated by “;” or spaces. N is the number of sizes.

1 KB = 1000 bytes; 1 MB = 10⁶ bytes.

Assumptions

Include TCP connection setup (1 RTT; the first request travels in the 3rd segment, with the ACK)

Server t_proc

View

Show

Client side

Highlight when the server→client link is idle Same time scale in both diagrams

Time unit

Height

Practice

Generates random values and hides the diagrams and results: draw both diagrams on paper, compute T and check your answer.

Export

Always export in light colors

Model assumptions

A single link of rate R and propagation delay d_prop between client and server; no other traffic, no losses, no TCP congestion control (no slow start).
Requests (GET) and TCP control segments have negligible size: they take d_prop to arrive and no transmission time.
Each object (and the HTML) is sent as a block of L bits: X = L/R. Headers are neglected.
Processing in the browser is neglected; the server needs t_proc to prepare each response.
Without pipelining, the client sends the request for object i when the last bit of object i−1 arrives.
With pipelining, the client sends all N requests at once; the server answers them in the same order, one after another. In the diagram the requests are drawn slightly apart only so they can be told apart.