SPDY is not so fast on real sites
Google conducted SPDY tests, which showed an increase in web page loading speed twice , and in mobile networks - by 23% . However, some independent experts express doubts about the effectiveness of the protocol. Recently, a critical review of SPDY was published by developers from Opera Software. And now Google has been hit hard - Guy Podjarny, a performance testing specialist, the author of the Mobitest service , one of the leading developers of Akamai, has posted his test results . They look depressing for SPDY. It turned out that in most cases SPDY + HTTPS works quite a bit faster than HTTPS, but, on average, 3.4% slower than simple HTTP.
Method Fry such that not undermine. 500 sites from the Alexa top tag (the author says that the number of porn sites is scary there, but he didn’t exclude them from the sample) were downloaded via SPDY proxies, Cotendo CDN servers (recently acquired by Akamai) were used as proxies. Cotendo is one of the first content providers to implement SPDY support.
It should be noted that only original content of the site was placed on the proxy, and those files that were downloaded from third-party sites continued to be downloaded from third-party sites. This is a fundamental difference from the previous tests (apparently, even Google experts sinned with this not entirely honest optimization). The test showed that the average page requests resources from 18 different domains.
Chrome-agent WebPageTest was used as a client. In conventional Chrome browsers, SPDY support is randomly disabled in 5% of cases, as conceived by Google, and this makes testing difficult. In the case of the Chrome-agent WebPageTest, this function is deactivated, so that the SPDY runs continuously. At the time of testing WebPageTest corresponded to the version of Chrome 18.
')
Each web page was loaded five times on channels of different widths and with different delays. The entire test was repeated three times at different times of the day. In total, 90 thousand individual downloads were made, more than enough to minimize the statistical error.
results
Guy Podzharny explains the disastrous results of SPDY for several reasons. First, HTTP is not a bottleneck in the performance of most sites. Secondly, in modern websites, resources on web pages are loaded from different domains, including those with no SPDY support. Thirdly, SPDY does not solve the real problems with the speed of displaying web pages in the browser, which suffers from scripts and CSS.
Of course, if SPDY support is ubiquitous, and resources from third-party sites will also be loaded through SPDY, then the test results will improve. “SPDY is a terrific thing, it's the first real HTTP upgrade in more than ten years,” says Guy Podzharny.
Method Fry such that not undermine. 500 sites from the Alexa top tag (the author says that the number of porn sites is scary there, but he didn’t exclude them from the sample) were downloaded via SPDY proxies, Cotendo CDN servers (recently acquired by Akamai) were used as proxies. Cotendo is one of the first content providers to implement SPDY support.
It should be noted that only original content of the site was placed on the proxy, and those files that were downloaded from third-party sites continued to be downloaded from third-party sites. This is a fundamental difference from the previous tests (apparently, even Google experts sinned with this not entirely honest optimization). The test showed that the average page requests resources from 18 different domains.
Chrome-agent WebPageTest was used as a client. In conventional Chrome browsers, SPDY support is randomly disabled in 5% of cases, as conceived by Google, and this makes testing difficult. In the case of the Chrome-agent WebPageTest, this function is deactivated, so that the SPDY runs continuously. At the time of testing WebPageTest corresponded to the version of Chrome 18.
')
Each web page was loaded five times on channels of different widths and with different delays. The entire test was repeated three times at different times of the day. In total, 90 thousand individual downloads were made, more than enough to minimize the statistical error.
results
| Channel width (Down / Up Kbps, latency ms) | SPDY vs HTTPS | SPDY vs HTTP |
| Cable (5000/1000; 28) | SPDY 6.7% faster | SPDY is 4.3% slower |
| DSL (1500/384; 50) | SPDY 4.4% faster | SPDY 0.7% slower |
| Low latency mobile network (780/330; 50) | SPDY 3% faster | SPDY is 3.4% slower |
| High latency mobile network (780/330; 200) | SPDY is 3.7% faster | SPDY is 4.8% slower |
Of course, if SPDY support is ubiquitous, and resources from third-party sites will also be loaded through SPDY, then the test results will improve. “SPDY is a terrific thing, it's the first real HTTP upgrade in more than ten years,” says Guy Podzharny.
Source: https://habr.com/ru/post/145928/
All Articles