Speed Matters for Google Web Search Jake Brutlag Google, Inc. June 22, 2009 perceived differently by users due to the degree of partial rendering on the page. All other things being equal, more usage, as measured by number of searches, reflects more satisfied users. Table 1 gives the average daily searches per user over the experiment duration for the experiment group relative to the control group.
Google runs experiments on search traffic to understand and improve the search experience. A series of such experiments injected different types of server-side delay into the search results page load in order to understand the impact of latency on user behavior. In a given experiment, one group of users experienced the delay, while a second group served as the control. Across the experiments, the type of delay, the magnitude of the delay, and experiment duration (number of weeks users subject to delay) varied. Speed as perceived by the end user is driven by multiple factors. These experiments model only one of them: additional server processing time. Figure 1 illustrates the three types of delay tested; the type of delay distinguishes when the delay occurs in the sequence of steps that constitute a page load.
Table 1: Experiment Impact on Daily Searches Per User Type of Delay Magnitude Duration Impact Pre-header 50 ms 4 weeks — Pre-header 100 ms 4 weeks −0.20% Post-header 200 ms 6 weeks −0.29% Post-header 400 ms 6 weeks −0.59% Post-ads 200 ms 4 weeks −0.30% Average impact over 4 or 6 weeks hides any trend over time. By focusing on the subset of users who were part of the experiment (or control group) from the beginning (as identified by a browser cookie), one can determine if there is such a trend. Figure 2 illustrates the trend for the two 6 week experiments.
Figure 1: Injecting Server-Side Delay
Figure 2: Impact of Post-header Delays Over Time
−0.8%
−0.6%
−0.4%
−0.2%
0%
200 ms delay 400 ms delay
actual trend
−1%
In Figure 1, time flows from left to right and diagonal lines represent network communication between the client and server. A user perceives load time as the duration from start of navigation until page complete. The page load may include optional activities (in blue), but every search includes an HTTP request for the page HTML, spanning the duration between GET and Last Byte. Server processing commences upon receipt of the HTTP GET. Google search leverages HTTP chunking, which means the page header can be transmitted (and perhaps rendered) even before the results are ready. The different types of delay tested are: pre-header pausing all server processing for X ms immediately upon receipt of the HTTP GET post-header pausing for X ms after transmitting the page header but before any further server processing is done post-ads pausing X ms mid-stream sending the results, after the sponsored links Both the post-header and post-ads delay may be masked by network conditions. That is, a slow or unreliable connection may mitigate any delay in server processing from the user’s perspective. All three types of delay may be
daily searches per user relative to control
0.2%
Abstract – Experiments demonstrate that increasing web search latency 100 to 400 ms reduces the daily number of searches per user by 0.2% to 0.6%. Furthermore, users do fewer searches the longer they are exposed. For longer delays, the loss of searches persists for a time even after latency returns to previous levels.
wk1
wk2
wk3
wk4
wk5
wk6
The change over time can be quantified by dividing the 6 week duration in half. The impact on daily searches per user during the second half is −0.36% and −0.74% for the 200 ms and 400 ms delay respectively, compared to −0.22% and −0.44% for the first half. Users do fewer searches the longer they are exposed to delay. Furthermore, observing these users for the 400 ms delay after we stop subjecting them to the delay, the rate of daily searches per user for the experiment is still −0.21% relative to the control (averaged over the 5 weeks after removal of the delay). For longer delays, the loss of searches persists for a time even after latency returns to previous levels.
22 Jun 2009 - web search latency 100 to 400 ms reduces the daily number of ... periments injected different types of server-side delay into the search results ... In Figure 1, time flows from left to right and diago- nal lines represent network communication between the client and server. A user perceives load time as the du-.
Bridgevine is focused on customer acquisition empowered by its proprietary. Acquisition and Merchandising Platform, AMP. AMP combines SEO, SEM, mobile,. Daily Deals, Internet radio, display, social, and affiliates. Perhaps the most intriguing and con
British Petroleum (BP) is one of the world's largest companies with operations in more ... energy business â serve more than 15 million customers each day. .... Google is a trademark of Google Inc. All other company and product names may be tradema
Apr 26, 2010 - 3. INDEXING RDF DATA. The index structures that need to be built for any par- ticular search ... simplicity, we will call this a horizontal index on the basis that RDF ... a way to implement a secondary sort on values by rewriting.
Get Started. 1. Create a Google+ page with a photo and engage your followers frequently with high quality content. 2. In the 'website' section of your page's 'about' section, make sure that you include the green. URL that appears in your search resul
Get Started. 1. Create a Google+ page with a photo and engage your followers frequently with high quality content. 2. In the 'website' section of your page's 'about' section, make sure that you include the green URL that appears in your search result
All rights reserved. Nielsen and the Nielsen logo are registered trademarks or trademarks of CZT/ACN Trademarks, LLC. Used with permission. Other product and service names are trademarks of their respective companies. Mobile Search MoMentS. UnderStan
Search Ads 360 gives you the workflow tools, robust reporting, and ... is a unified advertising and analytics platform that helps enterprise marketers make better.
specifications exist (e.g., Web Services Transaction (WS-Transaction)1, Web ... 1 dev2dev.bea.com/pub/a/2004/01/ws-transaction.html. ... of the traffic network.
There was a problem previewing this document. Retrying... Download. Connect more apps... Try one of the apps below to open or edit this item. SPEED WEB ...
Hashtags are now even more powerful. When users search for certain hashtags, they'll see a stream of related. Google+ posts on the right hand side of Search results. For example, if someone searches for #agt in Google Search, they'll see this: How it
18: Studio Visit: SEO. 17: Terry Haggerty: Angle ...... 19: Interview with Vera Cortês / Vera Cortês Art Agency / ARCO 2008 Madrid, Spain. 18: Dan Perjovschi: Stu ...
Feb 4, 2016 - class pro:Role in PRO, or of its sub-classes in SCORO: ⢠scoro:contact-person. ⢠scoro:data-creator. ⢠scoro:data-curator. ⢠scoro:data-manager. ⢠pro:distributor. ⢠pro:editor. ⢠scoro:funder. ⢠scoro:host-institution.
Jun 3, 2016 - Oil near USD50/bbl but industry players not excited ... should disconnect oil services players' stock price with oil price as ..... Software Technology ⢠Telcos ..... constituting legal, accounting or tax advice, and that for accurate
Jun 3, 2016 - stronger confidence on oil price sustainability, there is little hope for a .... year, the sentiment from oil companies remains negative and capital .... Automotive ⢠Semiconductor ⢠Technology ..... Structured securities are comple
18: Studio Visit: SEO. 17: Terry Haggerty: Angle of Response / Kuttner Siebert Gallery, Berlin. 14: Interview with Dan Perjovschi at Fumetto Festival Lucerne.
10: Urs Fischer: Service à la Française (2009) / Luma Westbau / Pool etc. ...... 10: Claes Oldenburg & Coosje van Bruggen: The European Desktop / Ivorypress ...
Feb 4, 2016 - Changes the examples used for 6 Subject, and for 11 AlternateIdentifier. 5. Corrected an RDF term duplication in 7.2 contributorName. 6. Improvement to the formatting of the exemplar RDF statements, to enhance clarity. 7. Added âdata
âWe realized we were at a tipping point â we would have to pay more in licenses for our CMS to support the two external sites, and we saw that we could save IT ...
Nov 6, 2009 - Playstation 3" and ps3" were not syn- onyms twenty years ago; snp newspaper" and snp online" carry the same query intent only after snpon- line.com was published. Thus a static synonym list is less desirable. In summary, synonym discove
Results are presented for three types of typi- cal web audio ... dio or video file, while ignoring the actual audio content. This paper is ... of storage space. (Chelba ...