google (goo'gul') n. 1. a number 1 followed by 100 zeros  2. the worlds best search engine.

Google Briefs

Google's core technology conducts more than 200 million searches per day.

They have built the web's largest index of web pages (more than 3 billion).

Google is powered by the world's largest commercial Linux cluster (more than 10,000 servers).

The importance of Google cannot be overstated. Given the integration among the major portals and search engines, Google performs a substantial portion of all searches conducted on the Internet. A fair estimation of Google's contribution to a properly optimized site is in the 85% range (i.e. 85% of a site's traffic arising from an Internet search will originate from Google).

Google runs on a unique combination of advanced hardware and software. The heart of the software is PageRank (PR); the means of obtaining PageRank is accomplished by a number of different spiders/crawls (e.g. algorithms). It is worth noting that PageRank is not the sole determiner to actual rank. There are many instances of sites with lower PR scores ranking above sites with higher PR. Meaning 'more' is involved in the final rank determination. It is however, a significant contributor and as the algorithm becomes more sophisticated, will become increasingly more important.

Google's PageRank

Google's PageRank (PR) represents a citation graph developed from maps of a site's hyperlinks, page title and citations. It is a 'Citation Importance Ranking' system, defining a citation as an inbound link (backlink), resulting in an approximation of a page's importance and quality.

The actual PageRank Model:

number of outgoing links on page i

PageRank of page j (Idealized Model)

PageRank of page j (Actual Model)

A simplified look of the above:

We assume page A has pages p1...pn which point to it (i.e., they are the citations, backlinks, or inbound links). The parameter d is a damping factor which can be set between 0 and 1 (believed to be set at 0.85). The PageRank of a page A is given as follows:

PR of 'A' = (1-d) + d x [ PR(p1) / OL(p1) + PR(p2) / OL(p2) + ... PR(pn) / OL(pn) ]

again,

A represents a specific page
PR(pn) is the PageRank of the given page 'n'
OL(pn) is the number of outbound links on page 'n'
d is a constant (called the damping factor) which is approximately 0.85

Therefore,

PR of 'A' is (1 - 0.85) + (0.85) x (the summation of [the linking pages PR divided by the number of outbound links on the linking page] for all the pages linking to page 'A'

Thus, we can conclude the following:

1. PageRank is increased by the number of pages that link to it.
2. The higher the PageRank of a linking page the better.
3. The number of outbound links contained on the linking page has a diluting effect on its contribution.

Further contemplation on the equation that calculates PageRank results in the link structure strategy discussed previously.

Google's Different Crawls

Google does two types of crawls, a main crawl and a fresh crawl. The main crawl is done once a month and the fresh crawl is done constantly. Google is consistently altering these crawls, resulting in variations on which sites are crawled and how deep the selected sites are crawled (how many pages are spidered).

The monthly main crawl is a more comprehensive crawl then the fresh crawl. It visits all pages within the index and will crawl deeper. In essence, the main crawl is setting the stage for the big 'dance'.

The fresh crawl re-spiders pages that are already indexed and will also spider any new pages it finds along the way. New pages that are evaluated can be inserted into search results immediately, however the rank is not stable (the pages will actually 'disappear'). Do not let this short-term 'fresh gift' fool you, the pages have not yet been included into the main index. This observation can also be viewed with pages that are already in Google's main index but upon 'fresh' evaluation are re-ranked reflecting any new page changes.

The volatility of a 'fresh' rank will continue to occur until the page is properly added to the index. This occurs when the site has been through a main crawl and at least one 'dance'. The Google 'dance' generally occurs once a month (although at the time of this writing, the frequency as been extended to every 6 weeks). During the dance, pages are evaluated using PageRank. As explained above, PageRank of a particular page is dependent upon all the inbound links and the corresponding PageRanks.

Therefore, in order to calculate all the inbound links to a site, all pages in the index have to be evaluated first.

Think about this… How can you evaluate a part of a whole when the evaluation is dependent upon the evaluation of the whole, which is influenced, by the evaluation of the part?

Logically, this is not possible.

Therefore, a page's PageRank is always based on incomplete data (by definition of the PageRank algorithm, a PR score is only an approximation). Increased accuracy of Google's approximation results from running many iterations of the evaluation over the entire index (40 - 50 cycles). It is during this time that one will see specific pages jump up and down in rank (i.e. 'dance').

Completion of the dance (several days) concludes with new PR scores and updated backlinks (inbound) for all pages within the index, the so-called 'update'.