[tor-dev] Proposal: Optimistic Data for Tor: Client Side

[Kevin Bauer]

> Assuming you mean "stream" instead of "circuit" here, then, as above, I
> think most HTTP connections would be in this category.  It might be
> interesting to examine some HTTP traces to see, though.  <shoutout
> target="Kevin">Kevin, you were looking at some HTTP traces for other
> reasons, right?  Anything in there that may help answer this
> question?</shoutout>


This is a great question. 

Google released a study about a year ago [1] that characterized crawled web pages in terms of their total size, the number of distinct destination hosts per page, the number of HTTP GETs per page, and other attributes. Their data indicates that the median web page requires that the client connect to 5 distinct destination hosts and issue 6.25 GETs per host. 

Put another way, a typical web page requires 5 streams, each of which issue multiple GETs. The full distributions of these statistics are available at [1].

Assuming persistent HTTP connections, with this proposal, only a stream's initial GET request would experience an improvement in time-to-first-byte, while subsequent GETs would be unaffected. 

However, by reducing the time-to-first-byte for even just the first request per stream, the user is able to start fetching subsequent streams sooner, and thus, retrieve the whole page faster. 

Kevin

[1] https://code.google.com/speed/articles/web-metrics.html