- From: Richard Newman <rnewman@twinql.com>
- Date: Fri, 17 Apr 2009 14:57:19 -0700
- To: Paul Gearon <gearon@ieee.org>
- Cc: Juan Sequeda <juanfederico@gmail.com>, Semantic Web <semantic-web@w3.org>
>> Are Long URLs wasting bandwith? >> http://tech.slashdot.org/article.pl?sid=09/03/27/2017250&from=rss > > It's been a few years since I went as far down the protocol stack as > TCP/IP, but don't TCP packets have a minimum size? I would have > thought that most HTTP requests would fit into a single packet, with > no difference to the packet size. Or is my memory faulty here? It depends on the underlying transport protocol. If your IP traffic is going over an ATM switch, your data is being broken down into 48-byte frames, in which case saving a few bytes in URL length can make a difference of a frame or two for every request. On the other hand, if every host en route is using Ethernet-size packets, on the order of 1500 bytes, then yes, HTTP requests will typically fit into a single packet. Perhaps more important is the inclusion of URLs in response bodies, not requests. I was intrigued by this, so I did a little digging. Slashdot's front page right now is 75KB, which is about 50 1500B Ethernet frames (which might well be fragmented and recombined across multiple protocol layers en route). I was surprised to see that about 10KB of that data (remaining size 65138 versus 75252 bytes) is composed of URLs in href, action, and src attributes. Put another way: *outside* of protocol headers, 13% of Slashdot's HTML traffic is sending URLs back and forth. Shorter URLs, both internally and on the web at large (because many links are external) could cut their bandwidth bill by a noticeable amount. Not that significant compared to minifying CSS, Javascript, and HTML, or turning on gzip encoding, but every little helps when you're serving millions of visitors a day. (You'd also see a benefit at scale if you switched from XML to JSON or s-expressions. Every long tag name makes a difference!) Those shorter URLs will also impact HTTP requests and responses, 302s, 303s, and so on. I could certainly imagine a measurable reduction in bandwidth usage by shortening URLs, particularly in formats that are URI-heavy, like readable RDF encodings. Whether that translates into a meaningful financial saving is another matter, which seems to be the most common objection to this article. -R
Received on Friday, 17 April 2009 21:57:57 UTC