The Anatomy of a Large Scale Web Search Engine

Abstract

Global search engines are an integral part of the World Wide Web. They have made the rapid growth of the web possible by allowing users to find web pages relevant to their interests in a sea of information.

However, to engineer a search engine is a challenging task. Search engines index tens to hundreds of millions of web pages involving a comparable number of distinct terms. They answer tens of millions of queries every day. Apart from the problems of scaling traditional search techniques to data of this magnitude, there are new technical challenges because of the demands of uncontrolled hypertext collections which are only starting to be met.

Web search engines are very different from traditional search engines in that they operate on hypertext. Therefore, they have to deal with crawling and can make use of links for searching. Furthermore, due to rapid advance in technology and web proliferation, creating a web search engine today is very different from three years ago.

In this paper, we present Google, a prototype of a large scale search engine. Google is designed to crawl and index the Web efficiently, using limited disk storage. To address scalability in search, Google makes use of hypertextual information found in links to produce higher quality results. The prototype with a full text of 24 million pages is available at http://google.stanford.edu/

1. Introduction

Since 1993, the World Wide Web has grown incredibly by almost any measure, from 130 web servers in June 1993 to 650,000 web servers in January 1997 [?] serving 100 million web pages. By making all of this information easy for users to locate, search engines have played a critical role in allowing the Web to scale to its present size. In this paper, we address the scalability of search engines in terms of both performance and search quality. We chose our system name, Google, because it is a common spelling of googol, or10100 and fits well with our goal of building very large scalable systems.

1.1 Web Search Engines: 1994 - 2000

Search engine technology has had to scale dramatically to keep up with the growth of the web. One of the first web search engines, the World Wide Web Worm (WWWW) [McBryan] had an index of 110,000 web pages and web accessible documents in 1994. As of November, 1997, the top search engines claim to index from 2 million (WebCrawler) to 100 million web documents [www.searchenginewatch.com]. It is foreseeable that by the year 2000, a comprehensive index of the WWW will contain over a billion documents. At the same time, the number of queries search engines handle has grown incredibly too. In March and April 1994, the World Wide Web Worm received an average of about 1500 queries per day. In November 1997, Altavista claimed it handled roughly 20 million queries per day. With the increasing number of users on the web, and automated systems which query search engines, it is likely that top search engines will handle hundreds of millions of queries per day by the year 2000. In this paper, we present the Google search engine developed at Stanford. The goal of Google is to address many of the problems introduced by scaling search engine technology to such extraordinary numbers.

1.2. Scaling with the Web

Creating a search engine which scales even to today's web presents many challenges. First, there are the performance considerations. It is necessary to have fast crawling technology to gather the web documents and keep them up to date. Storage space must be used efficiently to store indices and, optionally, the documents themselves. The indexing system must process hundreds of gigabytes of data efficiently. Queries must be handled quickly, at a rate of hundreds to thousands per second.

All of these tasks are becoming increasingly difficult as the Web grows. However, hardware performance and cost has also been improving dramatically to partially offset the difficulty. Disk cost has fallen to below $50 per gigabyte with transfer rates close to 10MB per second. Memory prices are below $5 per megabyte and 300MHz CPUs are available for little cost. There are, however, several notable exceptions this progress. Disk seek times have remained fairly high at about 10 ms because of the physical limitations of moving a disk head. To illustrate, it is now possible to read 100K in roughly the same time as performing one disk seek. Another problem is that operating system robustness still leaves much to be desired; operations on gigabytes of data often corrupt small portions of the data.

In designing Google, we have considered both the rate of growth of the WWW and technological changes. Google is designed to be scalable to extremely large data sets. It makes efficient use of storage space to store the index. The other data structures are optimized to allow for fast and efficient access (see section XXX). Further, we expect that the cost to index and store text or HTML is declining relative to the amount that will be available (see section XXX). This will result in favorable scaling properties for systems like Google.

1.3 Design Goals

Our primary design goal was to build an architecture that can support novel research activities on large scale web data. Another important design goal was to build systems that reasonable numbers of people can actually use. Usage was important to us because we think some of the most interesting research will involve leveraging the vast amount of usage data that is available from modern web systems. For example, there are many tens of millions of searches performed every day. However, it is very difficult to get this data, mainly because it is considered commercially valuable (see section XXX).

To support novel research uses, Google stores all of the actual documents it crawls in compressed form. One of our main goals in designing google was to set up an environment where other researchers can come in quickly and process large chunks of the web and produce interesting results that would have been very difficult otherwise. In the short time the system has been up, there have already been several papers using databases generated by Google, and about half a dozen others are underway. Indeed, our PageRank algorithm, described in greater detail below, would have been very difficult to build and evaluate without access to large chunks of the link structure of the web. Other projects using Google include "Dynamic Data Mining" [dynamic data mining reference], shiva, junghoo. One goal of Google is to set up a Spacelab like environment where researchers can propose and do interesting experiments on our systems and data.

1.3.1 Quality of Search

Besides supporting varied research, we have a strong goal to improve the quality of web search engines. In 1994, some people believed that a complete search index would make it possible to find anything easily. According to [Best of the Web 1994 - Navigators], ``The best navigation service should make it easy to find almost anything on the Web (once all the data is entered).'' However, the WWW of 1997 is quite different. Anyone who has used a search engine recently, can readily testify that the completeness of the index is not the only factor in the quality search results. ``Junk results'' often wash out any results that a user is interested in. In fact, as of November 1997, only one of the top four commercial search engines finds itself (returns its own search page in response to its name in the top ten results)  One of the main causes of this problem is that the number of documents in the indexes has been increasing many orders of magnitude, but the user's ability to look at documents has not. People can still only be willing to look at the first few tens of results. Because of this, as the collection size grows, we need tools that have high precision (number of relevant documents returned, say in the top tens of results) even at the expense of recall (the total number of relevant documents the system is able to return). There is no end in sight to this problem, as the number of documents on the web is still growing very rapidly. However, there is optimism that the use of more hypertextual information will help return better results. [two papers from www 97]. In particular, link structure [XXX?] and link text provide a lot of information for making relevance judgments and quality filtering. Google makes use of both link structure and anchor text (see Sections ?? and ??).

1.3.2 Academic vs. Commercial Search Engines

Aside from tremendous growth, WWW has also become increasingly commercial over time. In 1993, 1.5% of web servers were on .com domains. This number grew to over 60% in 1997. At the same time, search engines have migrated from the academic domain to the commercial. Currently most search engine development has gone on at companies with little publication of technical details. This causes search engine technology to remain largely a black art and to be advertising oriented (see Section ?). With Google, we have a strong goal to push more development and understanding into the academic realm.

2. System Features

The Google search engine has several features that help it produce better quality results. First, it makes use of link structure of the Web to calculate a quality ranking for each web page. This ranking is called PageRank. Second, Google treats link text specially for search purposes.

2.1 PageRank: Bringing Order to the Web

The web creates new challenges for information retrieval. The amount of information on the web is growing rapidly, as well as the number of new users inexperienced in the art of web research. People are likely to surf the web using its link graph, often starting with high quality human maintained indexes such as "YahooXXX!" or with search engines. Human maintained lists cover popular topics effectively but are subjective, expensive to build and maintain, slow to improve, and cannot cover all esoteric topics. Automated search engines that rely on keyword matching usually return too many low quality matches. To make matters worse, some advertisers attempt to gain people's attention by taking measures meant to mislead or "spam" automated search engines.

2.1.1 PageRank: Provides Part of the Solution

The citation graph of the web is an important resource that has largely gone unused in existing web search engines. We have created maps containing as many as 518 million of these hyperlinks, a significant sample of the total. These maps allow rapid calculation of a web page's "PageRank", an objective measure of its citation importance that corresponds well with people's subjective idea of importance. Because of this correspondence, PageRank is an excellent way to prioritize the results of web keyword searches. For most popular subjects, a simple text matching search that is restricted to web page titles performs admirably when PageRank prioritizes the results (demo available at google.stanford.edu). For the type of full text searches in the main Google system, PageRank also helps a great deal.

2.1.2 Definition of PageRank

Academic citation literature has been applied to the web, largely by counting citations or backlinks to a given page. This gives some approximation of a page's importance or quality. PageRank extends this idea by not counting links from all pages equally, and by normalizing by the number of links on a page.

We assume page A has pages T1...Tn which point to it (are citations). The factor d is a damping factor usually about 0.85, and is explained in the next section. Also C(A) is defined as the number of links going out of page A. The PageRank of a page A is given as follows:

PR(A) = (1-d) + d (PR(T1)/C(T1) + ... + PR(Tn)/C(Tn))

PageRank or PR(A) can be calculated using a simple iterative algorithm, and corresponds to the principal eigenvector of the normalized link matrix of the web. Also, a PageRank for every web page can be computed in few hours on a medium size workstation.

2.1.3 Intuitive Justification

A page can have a high PageRank if there are many pages that point to it or if the few pages that point to it have a high PageRank. This is useful, because pages that are well cited from many places around the web are worth looking at. Also, pages that have perhaps only one citations from something like the Yahoo! homepage are also worth looking at. If a page was not high quality, or was a broken link, it is quite likely that the Yahoo! home page would not link to it. PageRank handles both these cases and everything in between by recursively propagating weights through the link structure of the web:

PageRank can also be thought of as a model of user behavior. We assume there is a "random surfer" who is given a web page at random and keeps clicking on links, never hitting "back" but eventually gets bored and starts on another random page. The probability that the random surfer visits a page is its PageRank. And, the d damping factor is the probability at each page the "random surfer" will get bored and request another random page. We have several extensions to PageRank, but it is beyond the scope of this paper to discuss them here.

2.2 Anchor Text

The text of links is treated in a special way in our search engine. Normally, the text of a link gets associated with the page that the link is on. Instead, we associate it with both the page the link is on and the page the link points to. This has several advantages. First, anchors often provide more accurate descriptions of web pages than the pages themselves (see Section ??). Second, anchors may exist for documents which cannot be indexed by a text- based search engine such as images, programs, and databases. Third, this makes it possible to return web pages which have not actually been crawled. Note, there are problems with this approach, since pages are never checked for validity before being returned to the user. Because of this, it is possible that the search engine will return a page that never actually existed, but had hyperlinks pointing to it. However, because of the way that results are sorted, this problem happens rarely.

This idea of propagating anchor text to the page it refers to was implemented in the World Wide Web Worm [refXXXX] especially because it helps search non-text information, and expands the search coverage with fewer downloaded documents. We use it mostly for the first reason which is that anchor text can help provide better quality results. Using anchor text efficiently is technically difficult because of the large amounts of data which must be processed.  In our current crawl of 24 million pages, we had over 259 million anchors which we indexed.

2.3 Other Features

Aside from PageRank and the use of anchor text, Google has several other features. First, it has location information for all hits and so it makes extensive use of proximity in search. Second, Google keeps track of some visual presentation details such as font size of words. Words in a larger or bolder font are weighted higher than other words. Third, full raw HTML of pages is available in a repository.

3 Related Work

Search research on the web has a short and concise history. The WWWW was one of the first web search engines. It was subsequently followed by WebCrawler, Lycos, Altavista, Infoseek, Excite, HotBot (Inktome), and others. Of these more recent search engines, little has been published [Mauldin, Pinkerton]. Compared to the growth of the Web and the importance of search engines there are precious few documents about these search engines. According to Michael Mauldin (chief scientist, Lycos Inc) [Mauldin], "the various services (including Lycos) closely guard the details of these databases". However, there has been a fair amount of work on specific features of search engines. Especially well represented is work which can get results by post-processing the results of existing commercial search engines, or produce small scale "individualized" search engines. Finally, there has been a lot of research on information retrieval systems, especially on well controlled collections. In the next two sections, we discuss some areas where this research needs to be extended to work better on the web.

3.1 Information Retrieval

Work in information retrieval systems goes back many years and is well developed. However, most of the research on information retrieval systems is on small well controlled homogeneous collections such as collections of scientific papers or news stories on a related topic. Indeed, the primary benchmark for information retrieval, the Text Retrieval Conference [TREC 96], uses a fairly small, well controlled collection for their benchmarks. The "Very Large Corpus" benchmark is only 20GB compared to the 147GB from our crawl of 24 million web pages. Things that work well on TREC often do not produce good results on the web. For example, the standard vector space model tries to return the document that most closely approximates the query, given that both query and document are vectors defined by their word occurrence. On the web, this strategy often returns very short documents that are the query plus a few words. For example, we have seen a major search engine return a page containing only "Bill Clinton Sucks" and picture from a "Bill Clinton" query. Some argue that on the web, users should specify more accurately what they want and add more words to their query. We disagree vehemently with this position. If a user issues a query like "Bill Clinton" they should get reasonable results since there is a enormous amount of high quality information available on this topic. Given examples like these, we believe that the standard information retrieval work needs to be extended to deal effectively with the web.

3.2 Differences Between the Web and Well Controlled Collections

The web is a vast collection of completely uncontrolled heterogeneous documents. Documents on the web have extreme variation internal to the documents, and also in the external meta information that might be available. For example, documents differ internally in their language (both human and programming), vocabulary (email addresses, links, zip codes, phone numbers, product numbers), type or format (text, HTML, PDF, images, sounds), and may even be machine generated (log files or output from a database). On the other hand, we define external meta information as information that can be inferred about a document, but is not contained within it. Examples of external meta information include things like reputation of the source, update frequency, quality, popularity or usage, and citations. Not only are the possible sources of external meta information varied, but the things that are being measured vary many orders of magnitude as well. For example, compare the usage information from a major home page, like Yahoo's which currently receives XX million page views every day with an obscure historical article which might receive one view every ten years, differing by a huge order of magnitude. Clearly, these two items must be treated very differently by a search engine.

Another big difference between the web and traditional well controlled collections is that there is virtually no control over what people can put on the web. Couple this flexibility to publish anything with the enormous influence of search engines to route traffic and "fooling" or spamming search engines becomes a serious problem. There is enormous economic incentive to mislead search engines. If you can convince a search engine that your page should be returned for a popular query, even if you page has nothing to do with that query, you can reap enormous economic benefit because a large number of people will see your page; it is like free advertising. Because of this economic feedback system, people are willing to spend huge amounts of time tailoring their pages for search engines so they come up high in the results for important search terms. If the search engine keeps its index up to date, the problem is worsened since people have more feedback to adjust their pages. Finally, it is hard to keep track of offending parties, since they can easily create a new identity on the net. This "spamming" or misleading of search engines is a problem that has not been addressed in traditional information retrieval systems. Also, it is interesting to note that metadata efforts have largely failed with web search engines, because any text on the page which is not directly represented to the user is abused to "spam" search engines. There are even numerous companies which specialize in manipulating search engines for profit.

4 System Overview

A web search engine must perform several major functions: crawling, indexing, and searching. In this section we describe at a high level how each of these processes happens in Google. More detailed descriptions are in following sections.


Google Architecture Overview

Crawling is the most fragile task since it involves interacting with hundreds of thousands of web servers and various name servers which are all beyond the control of the system. Polite robot policies mandate that no server should be visited unreasonably often and the robots exclusion protocol should be heeded. Any errors in crawling invariably lead to many angry emails from webmasters. In fact, even polite behavior leads to some angry or confused emails from webmasters (see Section [?]). One way we have mitigated some of these problems is by only crawling sites that look like they are in the US (.com, .edu...). This reduces the number of people our system is in contact with, and gives us a denser sample.

In Google, the crawling is done by several distributed crawlers. There is a URLserver that sends lists of URLs to be fetched to the crawlers. The web pages that are fetched are then sent to the storeserver. The storeserver then compresses and stores the web pages into a repository. Every web page has an associated ID number called a docID which is assigned whenever a new URL is parsed out of a web page.  The indexing function is performed by the indexer and the sorter. The indexer performs a number of functions. It reads the repository, uncompresses the documents, and parses them. Each document is converted into a set of word occurrences called hits. The hits record the word, position in document, an approximation of font size, and capitalization. The indexer distributes these hits into a set of ``barrels'', creating a partially sorted forward index.

The indexer performs another important function. It parses out all the links in every web page and stores important information about them in an anchors file. This file contains enough information to determine where each link points from and to and the text of the link.

The URLresolver read the anchors file and converts relative urls into absolute urls and in turn into docids. It puts the anchor text into the forward index, associated with the docid that the anchor points to. It also generates a database of links which are pairs of docids. The links database is used to compute PageRanks for all the documents.

The sorter takes the barrels, which are sorted by docID (this is a simplification, see Section ??), and resorts them by wordID to generate the inverted index. This is done in place so that little temporary space is needed for this operation. The sorter also produces a list of wordids and offsets into the inverted index. A program called dumplexicon takes this list together with the lexicon produced by the indexer and generates a new lexicon to be used by the searcher. The searcher is run by a web server and uses the lexicon built by dumplexicon together with the inverted index and the PageRanks to answer queries.

Major Data Structures

Google is composed of a number of important data structures.  These data structures are optimized so that a large document collection can be crawled, indexed, and searched with little cost.  They are designed to scale well with large amounts of data and make use of technological changes that have happened in hardware. Although, CPUs and bulk input output rates have improved dramatically over the years, a disk seek still requires about 10 ms to complete.  Therefore, it would take three days to perform a seek for every document in our current 24 million page collection.  This is not completely unreasonable and further speedup is possible through parallelizing disk seeks. However, seeks are a major bottleneck and it seems unlikely that seek performance will improve as rapidly as other parameters assuming traditional disk technology.  Therefore, Google is designed to avoid disk seeks whenever possible, and does not even perform one disk seek for every document.

BigFiles

Many of Google's data structures are often larger than 4 gigabytes (the maximum file size addressable by a 32 bit integer) and they must be spread among drives.  We considered using a database but chose to implement our own bigfile data structure for portability, efficiency, and fine level control.  Currently BigFiles are virtual files spanning multiple file systems and are addressable by 64 bit integers.  The allocation among multiple file systems is handled automatically.  BigFiles also support rudimentary compression options. We plan to extend bigfiles to inherently support distributed processing by allowing multiple readers and writers.

Repository

The repository contains the full HTML of every web page.  Each page is compressed using zlib (see RFC 1950). The choice of compression technique is a tradeoff between speed and compression ratio. We chose zlib's speed over a significant improvement in compression offered by bzip. The compression rate of bzip was approximately 4 to 1 on the repository as compared to zlib's 3 to 1 compression. In the repository, the documents are stored one after the other and are prefixed by docid, length, and url. The repository requires no other data structures to be used in order to access it. This helps with data consistency and make development much easier; we can rebuild all the other data structures from only the repository and a file which lists crawler errors.

Document Index

The document index keeps information about each document.  It is a fixed width ISAM (Index sequential access mode) index, ordered by docid.  The information stored in each entry includes the current document status (reference seen, sent to urlserver, crawled, ...), a pointer into the repository, a document checksum, and various statistics (number of words, number of question marks, last crawl date, ...).  If the document has been crawled, it also contains a pointer into a variable width file called docinfo which contains its URL and title.  Otherwise the pointer points into the URLlist which contains just the URL. This design decision was driven by the desire to have a reasonably compact data structure, and the ability to fetch a record in one disk seek. The compactness is important since our list of 76.5 million known URLs generated by crawling 24 million URLs takes over 4GB of space with no titles or empty space.

Additionally, there is a file which is used to convert URLs into docIDs.  It is a list of URL checksums with their corresponding docIDs and is sorted by checksum.  In order to find the docID of a particular URL, the URL's checksum is computed and a binary search is performed on the checksums file to find its docid.  URLs may be converted into docIDs in batch by doing a merge with this file.  This is the technique the URLresolver uses to turn URLs into docids.  This batch mode of update is crucial because otherwise we must perform one seek for every link which would take more than one month for our 24 million page, 322 million link dataset.

Lexicon

The lexicon has several different forms.  One important change from earlier systems is that the lexicon can fit in memory for a reasonable price.  In the current implementation we can keep the lexicon in memory on a machine with 256 MB of main memory.  The current lexicon contains 14 million words (though some rare words were not added to the lexicon).  It is implemented in two parts -- a list of the words (concatenated together but separated by nulls) and a hash table of pointers.  For various functions, the list of words has some auxiliary information which is beyond the scope of this paper to explain fully.

Hit Lists

A hit list corresponds to a list of occurrences of a particular word in a particular document including position, font, and capitalization information. Hit lists account for most of the space used in both the forward and the inverted indices. Because of this, it is important to represent them as efficiently as possible. We considered several alternatives for encoding position, font, and capitalization - simple encoding (a triple of integers), a compact encoding (a hand optimized allocation of bits), and Huffman coding, In the end we chose a hand optimized compact encoding since it required far less space than the simple encoding and far less bit manipulation than Huffman coding.

Our compact encoding uses two bytes for every hit. There are two types of hits: fancy hits and plain hits. Fancy hits include hits occurring in a URL, title, anchor text, or meta tag;. plain hits include everything else. A plain hit consists of a capitalization bit, font size (relative to the rest of the document) in three bits (only 7 values are actually used because 111 is the flag that signals a fancy hit), and position as number of words in 12 bits (all positions higher than 4095 are labeled 4096). A fancy hit consists of a capitalization bit, the font size set to 7 to indicate it is a fancy hit, 4 bits to encode the type of fancy hit, and 8 bits of position. For anchor hits, the 8 bits of position are split into 4 bits for position in anchor and 4 bits for a hash of the docid the anchor occurs in. This gives us some limited phrase searching as long as there are not that many anchors for a particular word. We expect to update the way that anchor hits are stored to allow for greater resolution in the position and docid hash fields. We use font size relative to the rest of the document because when searching, you do not want to rank otherwise identical documents differently just because one of the documents is in a larger font.

The length of a hit list is stored before the hits themselves. To save space, the length of the hit list is combined with the wordid in the forward index and the docid in the inverted index. This limits it to 8 and 5 bits respectively (there are some tricks which allow 8 bits to be borrowed from the wordid). If the length is longer than would fit in that many bits, an escape code is used in those bits, and the next two bytes contain the actual length.


Barrels of Hits

Forward Index

The forward index is actually already partially sorted. It is stored in a number of barrels (we used 64). Each barrel holds a range of wordid's. If a document contains words that fall into a particular barrel, the docid is recorded into the barrel, followed by a list of wordid's with hitlists which correspond to those words. This scheme requires slightly higher storage because of duplicated docids but the difference is very small for a reasonable number of buckets and saves considerable time and coding complexity in the final indexing phase done by the sorter. Furthermore, instead of storing actual wordid's, we store each wordid as a relative difference from the minimum wordid that falls into the barrel the wordid is in. This way, we can use just 24 bits for the wordid's in the unsorted barrels, leaving 8 bits for the hit list length.

Inverted Index

The inverted index consists of the same barrels as the forward index, except that they have been processed by the sorter. For every valid wordid, the lexicon contains a pointer into the barrel that wordid falls into. It points to a doclist of docid's together with their corresponding hit lists. This doclist represents all the occurrences of that word in all documents.

An important issue is in what order the docid's should appear in the doclist. One simple solution is to store them sorted by docid. This allows for quick merging of different doclists for multiple word queries. Another option is to store them sorted by a ranking of the occurrence of the word in each document. This makes answering one word queries trivial and makes it likely that the answers to multiple word queries are near the start. However, merging is much more difficult. Also, this makes development much more difficult in that a change to the ranking function requires a rebuild of the index.

We chose a compromise between these options. That is to keep two sets of inverted barrels - one set for hit lists which include title or anchor hits and another set for all hit lists. This way, we check the first set of barrels first and if there are not enough matches within those barrels we check the larger ones. In our experiments, the title+anchor barrels have been roughly 10% the size of the corresponding full barrels.

Crawling the Web

Running a web crawler is a challenging task. There are tricky performance and reliability issues and even more importantly, there are social issues.

Efficient Crawling

In order to scale to hundreds of millions of web pages, Google has a fast distributed crawling system. A single urlserver serves lists of urls to a number of crawlers (we typically ran about 3). Both the urlserver and the crawlers are implemented in Python. Each crawler keeps roughly 300 connections open at once. This is necessary to retrieve web pages at a fast enough pace. At peak speeds, the system can crawl over 100 web pages per second using four crawlers. This amounts to roughly 600K per second of data. A major performance stress is DNS lookup. Each crawler maintains a its own DNS cache so it does not need to do a DNS lookup before crawling each document.

Each of the hundreds of connections can be in a number of different states: looking up DNS, connecting to host, sending request, and receiving response. Moreover, in order to be reasonably reliable, certain failures like DNS must be retried several times. These factors make the crawler a complex component of the system. It uses asynchronous IO to manage events, and a number of queues to move page fetches from state to state.

Crawler Reliability

The WWW is very heterogeneous, which is a delight to surfers who like variety but quite a burden on a program which must handle anything. In our crawls, we encountered infinite web pages, infinite URLs, many varied kinds of communication errors, and anything else one might imagine. As an amusing example, a number of hosts had their IP address resolve to 127.0.0.1 - the local host. As a result, during early runs, we were surprised how many web pages matched terms from our own home page. This happened because the crawler would crawl a site where the IP address resolved to 127.0.0.1 which means the local machine. Since there was a web server running on the crawler's machine, the crawler would crawl that server, but it would index it with the name of the site which for some reason resolved to 127.0.0.1.

Social Issues

It turns out that running a crawler which connects to more than half a million servers, and generates tens of millions of log entries generates a fair amount of email and phone calls. Because of the vast number of people coming on line, there are always those users who do not know what a crawler is, because this is the first one they have seen. Almost daily, we receive an email something like, "Wow, you looked at a lot of pages from my web site. How did you like it?" There are also some people who do not know about the robots exclusion protocol, and think their page should be protected from indexing by a statement like, "This page is copyrighted and should not be indexed", which needless to say is difficult for web crawlers to understand. Also, because of the huge amount of data involved, unexpected things will happen. For example, our system was trying to crawl an online game. This resulted in lots of garbage messages in the middle of their game! It turns out this was an easy problem to fix. But this problem had not come up until we had downloaded tens of millions of pages. Because of the immense variation in web pages and servers, it is virtually impossible to test a crawler without running it on large part of the Internet. Invariably, there are hundreds of obscure problems which may only occur on one page on the whole web and cause the crawler to crash, or worse, unpredictable or incorrect behavior.
 
Since such large numbers of people are looking at their web logs every day, if only one out of ten thousand people contact us we will be drowning in email. As a result, systems which access large parts of the Internet need to be designed to be very robust and carefully tested. Since large complex systems such as crawlers will invariably cause problems, there needs to be significant resources devoted to reading the email and dealing with these problems as they come up.

8 Indexing the Web

The whole indexing process involves three major steps: parsing, encoding, and sorting. All of these deal with a huge volume of data and therefore they must be tolerant of data corruption. Data can get corrupted in many ways including software bugs, operating system bugs, and hardware bugs. Of course, the data that the parser receives is raw web pages and those have errors for reasons of their own.

The Parser

Any parser which is designed to run on the entire Web must handle a huge array of possible errors. These range from typos in HTML tags to kilobytes of zeros in the middle of a tag, non-ascii characters, HTML tags nested hundreds deep, and a great variety of other errors that challenge anyone's imagination to come up with equally creative ones. In order to deal with these kinds of problems, our parser maintains its own stack. Instead of using yacc to generate a CFG parser, we use flex to generate a lexical analyzer which we outfit with its own stack.

Indexing Documents into Barrels

After each document is parsed, it is encoded into a number of barrels. Every word is converted into a wordid by using an in-memory hash table - the lexicon. New additions to the hash table are logged to a file. The main difficulty with parallelization of the indexing phase is that the lexicon must be shared. Once the words are converted into wordid's, their occurrences in the current document are translated into hit lists and are written into the forward barrels.

Sorting

In order to generate the inverted index, the sorter takes each of the forward barrels and sorts it by wordid to produce an inverted barrel for title and anchor hits and a full inverted barrel. This process happens one barrel at a time, thus requiring little temporary storage. Also, we parallelize the sorting phase to use as many machines as we have simply by running multiple sorters, which can process different buckets at the same time. It required roughly about a day of wall clock time to sort our 24 million page index on three machines with 256MB of RAM each (a significant amount of this time was spent doing slow file IO over NFS).

Since the barrels don't fit into main memory, the sorter further subdivides them into baskets which do fit into memory based on wordid and docid. Then the sorter, loads each basket into memory, sorts it and writes its contents into the short inverted barrel and the full inverted barrel.

Searching

The goal of searching is to provide quality search results efficiently. Many of the large commercial search engines seemed to have made great progress in terms of efficiency. Therefore, we have focused more on quality of search in our research.

Given a query Google evaluates it as follows:

  1. Parse the query.
  2. Convert words into wordids.
  3. Seek to the start of the doclist in the short barrel for every word.
  4. Scan through the doclists until there is a document that matches all the search terms.
  5. Compute the rank of that document for the query.
  6. If we are in the short barrels and at the end of any doclist, seek to the start of the doclist in the full barrel for every word and go to step 4.
  7. If we are not at the end of any doclist go to step 4.
  8. Sort the documents that have matched by rank and return the top k.

To put a limit on response time, once a certain number (currently 40000) of matching documents are found, the searcher automatically goes to step 8. This means that it is possible that suboptimal results would be returned. We are currently investigating other ways to solve this problem. In the past, we sorted the hits according to PageRank, which seemed to help.

The Ranking Function

Google maintains much more information about web documents than typical search engines. Every hitlist includes position, font, and capitalization information. Additionally, we factor in hits from anchor text and the PageRank of the document. Combining all of this information into a rank is difficult. We designed our ranking function so that no one factor can have too much influence.

First, consider the simplest case - a single word query. In order to rank a document with a single word query, Google looks at that document's hit list for that word. Google considers each hit to be one of several different types (title, anchor, url, plain text large font, plain text small font, ...), each of which has its own type-weight. The type-weights make up a vector indexed by type. Google counts the number of hits of each type in the hit list. Then every count is converted into a count-weight. Count-weights increase linearly with counts at first but quickly taper off so that more than a certain count will not help. We take the dot product of the vector of count-weights with the vector of type-weights to compute an IR score for the document. Finally, the IR score is combined with PageRank to give a final rank to the document.

For a multi-word search, the situation is more complicated. Now multiple hit lists must be scanned through at once so that hits occurring close together are weighted higher than hits occurring far apart. The hits from the multiple hit lists are matched up so that nearby hits are matched together. For every matched set of hits, a proximity is computed. The proximity is based on how far apart the hits are in the document (or anchor) but is binned into 10 different values ranging from a phrase match to "not even close".

Counts are computed not only for every type of hit but for every type and proximity. Every type and proximity pair has a type-prox-weight. The counts are converted into count-weights and we take the dot product of the count-weights and the type-prox-weights to compute an IR score.

Feedback

The ranking function has many parameters like the type-weights and the type-prox-weights. Figuring out the right values for these parameters is something of a black art. In order to do this, we have a user feedback mechanism in the search engine. A user may optionally evaluate all of the results that are returned. This feedback is saved. Then when we modify the ranking function, we can see the impact of this change on all previous searches which were ranked. This gives us some idea of how a change in the ranking function affects the search results.

Results

SHOW SOME SAMPLE RESULTS -- help me find some good ones if you want http://z.stanford.edu:4200/ (search the web)

13 Conclusions


Notes:

Future work?

Related work? Mention caching strategies instead of crawling

Distribute crawlers to multiple locations -- even have java version.

URLS

Bzip2 Home Page
http://www.muraroa.demon.co.u

The Effect of Cellular Phone Use Upon Driver Attention
http://www.webfirst.com/aaa/text/cell/cell0toc.htm

Search Engine Watch
http://www.searchenginewatch.com/

RFC 1950 (zlib)
ftp://ftp.uu.net/graphics/png/documents/zlib/zdoc-index.html

Robots Exclusion Protocol:
http://info.webcrawler.com/mak/projects/robots/exclusion.html

References

two papers from 97 web conference

[McBryan] WWWW Paper Webcrawler Description Rankdex Web Growth BOTW 1994 - Navigators

[Mauldin]

second 94 web conference Finding What People Want: Experiences with the WebCrawler Pinkerton

mg

[Bagdikian 97] Ben H. Bagdikian. The Media Monopoly. 5th Edition. Publisher: Beacon, ISBN: 0807061557

[Brin 98] Sergey Brin and Lawrence Page, Dynamic Data Mining: A New Architecture for Data with High Dimensionality. submitted to SIGMOD '98.

[Cho 98] Junghoo Cho, Hector Garcia-Molina, Lawrence Page. Efficient Crawling Through URL Ordering. Submitted to the Seventh International Web Conference (WWW 98). Brisbane, Australia, April 14-18, 1998.

[Gravano 94] Luis Gravano, Hector Garcia-Molina, and A. Tomasic. The Effectiveness of GlOSS for the Text-Database Discovery Problem. Proc. of the 1994 ACM SIGMOD International Conference On Management Of Data, 1994.

[Marchiori 1997] Massimo Marchiori. The Quest for Correct Information on the Web: Hyper Search Engines. The Sixth International WWW Conference (WWW 97). Santa Clara, USA, April 7-11, 1997.

[Page 98] Lawrence Page and Sergey Brin. PageRank, an Eigenvector based Ranking Approach for Hypertext. Submitted to the 21st Annual ACM/SIGIR International Conference on Research and Development in Information Retrieval. Melbourne, Australia, August 24 - 28, 1998.

[Pinkerton 1994] Brian Pinkerton, Finding What People Want: Experiences with the WebCrawler. The Second International WWW Conference Chicago, USA, October 17-20, 1994.

[Shivakumar 98] Narayanan Shivakumar, Hector Garcia-Molina, Chandra Chekuri. Filtering with Approximate Predicates. Submitted to SIGMOD '98.

[TREC 96] Proceedings of the fifth Text REtrieval Conference (TREC-5). Gaithersburg, Maryland, November 20-22, 1996. Publisher: Department of Commerce, National Institute of Standards and Technology. Editors: D. K. Harman and E. M. Voorhees. Full text at: http://trec.nist.gov/

@InProceedings{hyper96*180,
  author =       "Ron Weiss and Bienvenido V{\'e}lez and Mark A. Sheldon
                 and Chanathip Manprempre and Peter Szilagyi and Andrzej
                 Duda and David K. Gifford",
  title =        "{HyPursuit}: {A} Hierarchical Network Search Engine
                 that Exploits Content-Link Hypertext Clustering",
  pages =        "180--193",
  ISBN =         "0-89791-778-2",
  booktitle =    "Proceedings of the 7th {ACM} Conference on Hypertext",
  month =        "16--20~" # mar,
  publisher =    "ACM Press",
  address =      "New York",
  year =         "1996",
}
from www1.cern.ch/WWW94/PerlimProcs.html first web conference
Keywords: indexing, searching
     Title: GENVL and WWWW: Tools for Taming the Web
     Author: Oliver A. McBryan <mcbryan@cs.colorado.edu>
     Institute: University of Colorado, Boulder, CO, US
     PostScript, Size: 70866, Printed: 11 pages
  • Keywords: indexing, searching
    Title: GENVL and WWWW: Tools for Taming the Web
    Author: Oliver A. McBryan <mcbryan@cs.colorado.edu>
    Institute: University of Colorado, Boulder, CO, US
    PostScript, Size: 70866, Printed: 11 pages

  • Keywords: indexing, technology, search, http
    Title: Aliweb - Archie-Like Indexing in the Web
    Author: Martijn Koster <m.koster@nexor.co.uk>
    Institute: NEXOR Ltd., UK
    PostScript, Size: 213616, Printed: 10 pages

  • Keywords: technology, caching, servers
    Title: What can Archives offer the World Wide Web
    Author: Neil Smith <ngs@ukc.ac.uk>
    Institute: Unix Hensa, The University of Kent at Canterbury, UK
    PostScript, Size: 442678, Printed: 12 pages

  • Keywords: indexing, spiders, repositories
    Title: The RBSE Spider - Balancing Effective Search Against Web Load
    Author: David Eichmann <eichmann@rbse.jsc.nasa.gov>
    Institute: RICIS, University of Houston, Clear Lake, US
    PostScript, Size: 55372, Printed: 8 pages

  • Keywords: indexing, searching
    Title: Lost in Hyperspace? Free Text Searches in the Web
    Author: Christian Neuss <neuss@igd.fhg.de>
    Author: Stefanie Höfling <hoefling@igd.fhg.de>
    Institute: Fraunhofer Institute for Computer Graphics, Darmstadt, DE
    PostScript, Size: 987758, Printed: 8 pages

  • Keywords: indexing, searching, wais, integration, servers
    Title: Towards Better Integration of Dynamic Search Technology and the World-Wide Web
    Author: Doug McKee <doug@navisoft.com>
    Institute: Navisoft, US
    PostScript, Size: 43638, Printed: 7 pages

  • Keywords: indexing, searching, browsing, mosaic
    Title: Information Retrieval in the World-Wide Web: Making Client-based searching feasible
    Author: R.D.J. Post <reinpost@win.tue.nl>
    Author: P.M.E. De Bra <debra@win.tue.nl>
    Institute: Eindhoven University of Technology, NL
    PostScript, Size: 162778, Printed: 10 pages
  • Appendix A: Advertising and Mixed Motives

    Currently, the predominant business model for commercial search engines is advertising. The goals of the advertising business model do not always correspond to providing quality search to users. For example, in our prototype search engine the top result for cellular phone is "The Effect of Cellular Phone Use Upon Driver Attention", a study which explains in great detail the distractions and risk associated with conversing on a cell phone while driving. This search result came up first because of its high importance as judged by the PageRank algorithm, an approximation of citation importance on the web [Page, 98]. It is clear that a search engine which was taking money for showing cellular phone ads would have difficulty justifying the page that our system returned to its paying advertisers. For this type of reason and historical experience with other media [Bagdikian 83], we expect that advertising funded search engines will be inherently biased towards the advertisers and away from the needs of the consumers. Since it is very difficult even for experts to evaluate search engines, search engine bias is particularly insidious. A good example was OpenText, which was reported to be selling companies the right to be listed at the top of the search results for particular queries. This type of bias is much more insidious than advertising, because it is not clear who "deserves" to be there, and who is willing to pay money to be listed. This business model resulted in an uproar, and OpenText has ceased to be a viable search engine. But less blatant bias are likely to be tolerated by the market. For example, a search engine could add a small factor to search results from "friendly" companies, and subtract a factor from results from competitors. This type of bias is very difficult to detect but could still have a significant effect on the market. Furthermore, advertising income often provides an incentive to provide poor quality search results. For example, we noticed a major search engine would not return a large airline's home page when the airlineÕs name was given as a query. It so happened that the airline had placed an expensive ad, linked to the query that was its name. A better search engine would not have required this ad, and possibly resulted in the loss of the revenue from the airline. In general, it could be argued from the consumer point of view that the better the search engine is, the fewer advertisements will be needed for the consumer to find what they want. This of course erodes the advertising supported business model of the existing search engines. However, there will always be money from advertisers who want a customer to switch products, or have something that is genuinely new. But we believe the issue of advertising causes enough mixed incentives that it is crucial to have a competitive search engine that is transparent and in the academic realm. 

    Appendix B: Scalability 

    Scalability of Google

    We have designed Google to be scalable in the near term to a goal of 100 million web pages. We have disk and machines on the way to handle roughly that amount. All of the time consuming parts of the system are parallelizable and roughly linear time. These include things like the crawlers, indexers, and sorters. We also think that most of the data structures will deal gracefully with the expansion. However, at 100 million web pages we will be very close up against all sorts of operating system limits in the common operating systems (currently we run on both Solaris and Linux). These include things like addressable memory, number of open file descriptors, network sockets and bandwidth, and many others. We believe expanding to a lot more than 100 million pages would greatly increase the complexity of our system.

    Scalability of Centralized Indexing Architectures

    As the capabilities of computers have increased, it is possible to index an every larger amount of text for a reasonable cost. Of course, other more bandwidth intensive media such as video is likely to become more pervasive. But, because the cost of production of text is low compared to media like video, text is likely to remain very pervasive. Also, it is likely that soon we will have speech recognition that does a reasonable job converting speech into text, expanding the amount of text available. All of this provides amazing possibilities for centralized indexing. Here is an illustrative example. We assume we want to index everything everyone in the US has written for a year. We assume that there are 250 million people in the US and they write an average of 10k per day. That works out to be about 850 terabytes. Also assume that indexing a terrabyte can be done now for a reasonable cost. We also assume that the indexing methods used over the text are linear, or nearly linear in their complexity. Given all these assumptions we can compute how long it would take before we could index our 850 terrabytes for a reasonable cost assuming certain growth factors. Moore's Law was defined in 1965 as a doubling every 18 months in processor power. It has held remarkably true, not just for processors, but for other important system parameters such as disk as well. If we assume that Moore's law holds for the future, we need only 10 more doublings, or 15 years to reach our goal of indexing everything everyone in the US has written for a year for a price that a small company could afford. Of course, Moore's Law may not continue to hold, but there are certainly a lot of interesting centralized applications even if we only get part-way to our hypothetical example.

    Because humans can only type or speak a finite amount, and as computers continue improving, text indexing will scale even better than it does now. So we are optimistic that our centralized web search engine architecture will improve in its ability to cover the pertinent text information over time. Of course there will always be many problems where distributed systems like Gloss [Gravano 94] will be the best solution, but it often seems difficult to convince the world to use them. Distributed systems suffer in large part from high administration costs of setting up many systems, which is a problem that may be solved in the future.