22 - STUDENT - Instant Page Load (IPL)

Ivan Klimek (Technical University Kosice)

Probably the biggest lie in the history of Internet is that larger bandwidth directly translates to a better Web experience. The single factor of speed, or Mbps, became the de-facto “measure” of connection quality. In reality, it is not like that. For example, it was shown that an increase from 5Mbps to 10Mbps amounts to only 5 percent decrease in Page Load Times (PLT).
The real issue is the Round Trip Time (RTT), meaning the time it takes for a packet to get from the source to the destination and back. RTT is a physical hard limit set by the speed of light, so the only logical thing to do is to avoid doing exactly what is being done by the core protocol that the Web is built on – request each web page resource separately.
Hypertext Transport Protocol (HTTP) was designed back in the 1990s for simple web pages which contained only a few resources like images or scripts. Nowadays most pages have at least 40 resources. Current HTTP design creates a new TCP session for each requested page resource2 hence the TCP overhead due to handshakes and congestion control mechanisms is compounded.
Clearly HTTP is not best suited for what it is being used, but a protocol that the entire Internet world runs on cannot be easily replaced. Therefore there are only two possibilities: Either gradually replace HTTP by a more optimized protocol – an approach currently being taken by Google which is using its own in-house protocol SPDY for all its services by default on its Chrome browser (up to 64% decrease in PLT3) – or somehow transparently optimize the already deployed protocol, where transparently means without the need to change any part of the currently used server/client software.
This lighting talk will focus on the second mentioned approach. It will be the first public demonstration of our transparent network-based optimization system which was specifically developed to resolve the issues degrading HTTP performance, without the need to change a single web server nor client browser, while providing performance increase similar to the SPDY protocol. Both mentioned approaches are based on the same idea: Instead of downloading each resource separately they are downloaded in a single TCP stream thus the effect of RTT is limited when compared to HTTP. IPL is a network-based optimization system, it uses an automated Man-in-the-Middle attack to transparently intercept the client's HTTP request. Because it has a low RTT / large bandwidth Internet connection (preferably fiber) it can download the web page resources much faster than any client. The web page is then stripped of resources and a JavaScript script is added. This modified page is transparently returned to the client as a reply to its intercepted request. Once the page gets loaded, it creates a HTML5 WebSocket connection to the IPL device. This connection is then used
to download all the resources in a single stream. Lossless resource optimization (WebP/WebM conversion etc.) and/or caching may be also applied.

Download file