1. Thinking Differently About Web Page Preservation Michael L. Nelson, Frank McCown, Joan A. Smith Old Dominion University Norfolk VA {mln,fmccown,jsmit}@cs.odu.edu Library of Congress Brown Bag Seminar June 29, 2006 Research supported in part by NSF, Library of Congress and Andrew Mellon Foundation

2. Background “We can’t save everything!” –if not “everything”, then how much? –what does “save” mean?

3. “Women and Children First” image from: http://www.btinternet.com/~palmiped/Birkenhead.htmhttp://www.btinternet.com/~palmiped/Birkenhead.htm HMS Birkenhead, Cape Danger, 1852 638 passengers193 survivorsall 7 women & 13 children

4. We should probably save a copy of this…

5. Or maybe we don’t have to… the Wikipedia link is in the top 10, so we’re ok, right?

6. Surely we’re saving copies of this…

7. 2 copies in the UK 2 Dublin Core records That’s probably good enough…

8. What about the things that we know we don’t need to keep? You DO support recycling, right?

9. A higher moral calling for pack rats?

10. Just Keep the Important Stuff!

11. Lessons Learned from the AIHT images from: http://facweb.cs.depaul.edu/sgrais/collage.htm (Boring stuff: D-Lib Magazine, December 2005) Preservation metadata is like a David Hockney Polaroid collage: each image is both true and incomplete, and while the result is not faithful, it does capture the “essence”

12. Preservation: Fortress Model 1.Get a lot of $ 2.Buy a lot of disks, machines, tapes, etc. 3.Hire an army of staff 4.Load a small amount of data 5.“Look upon my archive ye Mighty, and despair!” image from: http://www.itunisie.com/tourisme/excursion/tabarka/images/fort.jpg Five Easy Steps for Preservation:

13. Alternate Models of Preservation Lazy Preservation –Let Google, IA et al. preserve your website Just-In-Time Preservation –Wait for it to disappear first, then a “good enough” version Shared Infrastructure Preservation –Push your content to sites that might preserve it Web Server Enhanced Preservation –Use Apache modules to create archival-ready resources image from: http://www.proex.ufes.br/arsm/knots_interlaced.htm

14. Lazy Preservation “How much preservation do I get if I do nothing?” Frank McCown

15. Web Infrastructure as a Resource Reconstructing Web Sites Research Focus Outline: Lazy Preservation

17. Web Infrastructure

19. Cost of Preservation H L H Publisher’s cost (time, equipment, knowledge) LOCKSS Browser cache TTApacheiPROXY Furl/Spurl InfoMonitor Filesystem backups Coverage of the Web H Client-view Server-view Web archives SE caches Hanzo:web

20. Web Infrastructure as a Resource Reconstructing Web Sites Research Focus Outline: Lazy Preservation

21. Research Questions How much digital preservation of websites is afforded by lazy preservation? –Can we reconstruct entire websites from the WI? –What factors contribute to the success of website reconstruction? –Can we predict how much of a lost website can be recovered? –How can the WI be utilized to provide preservation of server-side components?

22. Prior Work Is website reconstruction from WI feasible? –Web repository: G,M,Y,IA –Web-repository crawler: Warrick –Reconstructed 24 websites How long do search engines keep cached content after it is removed?

23. Timeline of SE Resource Acquisition and Release Vulnerable resource – not yet cached (t ca is not defined) Replicated resource – available on web server and SE cache (t ca < current time < t r ) Endangered resource – removed from web server but still cached (t ca < current time < t cr ) Unrecoverable resource – missing from web server and cache (t ca < t cr < current time) Joan A. Smith, Frank McCown, and Michael L. Nelson. Observed Web Robot Behavior on Decaying Web Subsites, D-Lib Magazine, 12(2), February 2006. Frank McCown, Joan A. Smith, Michael L. Nelson, and Johan Bollen. Reconstructing Websites for the Lazy Webmaster, Technical report, arXiv cs.IR/0512069, 2005.

26. Cached Image

27. Cached PDF http://www.fda.gov/cder/about/whatwedo/testtube.pdf MSN version Yahoo versionGoogle version canonical

28. Web Repository Characteristics TypeMIME typeTypical file ext GoogleYahooMSNIA HTML text text/html html CCCC Plain text text/plain txt, ans MMMC Graphic Interchange Format image/gif gif MM~RC Joint Photographic Experts Group image/jpeg jpg MM~RC Portable Network Graphic image/png png MM~RC Adobe Portable Document Format application/pdf pdf MMMC JavaScript application/javascript js MMC Microsoft Excel application/vnd.ms-excel xls M~SMC Microsoft PowerPoint application/vnd.ms-powerpoint ppt MMMC Microsoft Word application/msword doc MMMC PostScript application/postscript ps M~SC CCanonical version is stored MModified version is stored (modified images are thumbnails, all others are html conversions) ~RIndexed but not retrievable ~SIndexed but not stored

29. SE Caching Experiment Create html, pdf, and images Place files on 4 web servers Remove files on regular schedule Examine web server logs to determine when each page is crawled and by whom Query each search engine daily using unique identifier to see if they have cached the page or image Joan A. Smith, Frank McCown, and Michael L. Nelson. Observed Web Robot Behavior on Decaying Web Subsites. D-Lib Magazine, February 2006, 12(2)Observed Web Robot Behavior on Decaying Web Subsites

30. Caching of HTML Resources - mln

31. Reconstructing a Website Warrick Starting URL Web Repo Original URL Results page Cached URL Cached resource File system Retrieved resource 1.Pull resources from all web repositories 2.Strip off extra header and footer html 3.Store most recently cached version or canonical version 4.Parse html for links to other resources

32. How Much Did We Reconstruct? A “Lost” web site Reconstructed web site BC DEF A B’C’ GE F Missing link to D; points to old resource G F can’t be found

33. Reconstruction Diagram added 20% identical 50% changed 33% missing 17%

34. Websites to Reconstruct Reconstruct 24 sites in 3 categories: 1. small (1-150 resources) 2. medium (150-499 resources) 3. large (500+ resources) Use Wget to download current website Use Warrick to reconstruct Calculate reconstruction vector

35. Results Frank McCown, Joan A. Smith, Michael L. Nelson, and Johan Bollen. Reconstructing Websites for the Lazy Webmaster, Technical Report, arXiv cs.IR/0512069, 2005.Reconstructing Websites for the Lazy Webmaster,

36. Aggregation of Websites

37. Web Repository Contributions

38. Warrick Milestones www2006.org – first lost website reconstructed (Nov 2005)www2006.org DCkickball.org – first website someone else reconstructed without our help (late Jan 2006)DCkickball.org www.iclnet.org – first website we reconstructed for someone else (mid Mar 2006)www.iclnet.org Internet Archive officially “blesses” Warrick (mid Mar 2006) 1 1 http://frankmccown.blogspot.com/2006/03/warrick-is-gaining-traction.html http://frankmccown.blogspot.com/2006/03/warrick-is-gaining-traction.html

39. Web Infrastructure as a Resource Reconstructing Web Sites Research Focus Outline: Lazy Preservation

40. Proposed Work How lazy can we afford to be? –Find factors influencing success of website reconstruction from the WI –Perform search engine cache characterization Inject server-side components into WI for complete website reconstruction Improving the Warrick crawler –Evaluate different crawling policies Frank McCown and Michael L. Nelson, Evaluation of Crawling Policies for a Web-repository Crawler, ACM Hypertext 2006. –Development of web-repository API for inclusion in Warrick

41. Factors Influencing Website Recoverability from the WI Previous study did not find statistically significant relationship between recoverability and website size or PageRank Methodology –Sample large number of websites - dmoz.org –Perform several reconstructions over time using same policy –Download sites several times over time to capture change rates

42. Evaluation Use statistical analysis to test for the following factors: –Size –Makeup –Path depth –PageRank –Change rate Create a predictive model – how much of my lost website do I expect to get back?

43. Marshall TR Server – running EPrints

44. We can recover the missing page and PDF, but what about the services?

45. Recovery of Web Server Components Recovering the client-side representation is not enough to reconstruct a dynamically-produced website How can we inject the server-side functionality into the WI? Web repositories like HTML –Canonical versions stored by all web repos –Text-based –Comments can be inserted without changing appearance of page Injection: Use erasure codes to break a server file into chunks and insert the chunks into HTML comments of different pages

46. Recover Server File from WI

47. Evaluation Find the most efficient values for n and r (chunks created/recovered) Security –Develop simple mechanism for selecting files that can be injected into the WI –Address encryption issues Reconstruct an EPrints website with a few hundred resources

48. SE Cache Characterization Web characterization is an active field Search engine caches have never been characterized Methodology –Randomly sample URLs from four popular search engines: Google, MSN, Yahoo, Ask –Download cached version and live version from the Web –Examine HTTP headers and page content –Test for overlap with Internet Archive –Attempt to access various resource types (PDF, Word, PS, etc.) in each SE cache

49. Summary: Lazy Preservation When this work is completed, we will have… demonstrated and evaluated the lazy preservation technique provided a reference implementation characterized SE caching behavior provided a layer of abstraction on top of SE behavior (API) explored how much we store in the WI (server-side vs. client-side representations)

50. Web Server Enhanced Preservation “How much preservation do I get if I do just a little bit?” Joan A. Smith

51. OAI-PMH mod_oai: complex objects + resource harvesting Research Focus Outline: Web Server Enhanced Preservation

52. WWW and DL: Separate Worlds 1994 DL WWW Today The problem is not that the WWW doesn’t work; it clearly does. The problem is that our (preservation) expectations have been lowered. WWW DL “Crawlapalooza” “Harvester Home Companion”

53. “A repository is a network accessible server that can process the 6 OAI-PMH requests … A repository is managed by a data provider to expose metadata to harvesters.” “A harvester is a client application that issues OAI-PMH requests. A harvester is operated by a service provider as a means of collecting metadata from repositories.” Data Providers / Repositories Service Providers / Harvesters

54. Aggregators data providers (repositories) service providers (harvesters) aggregator aggregators allow for: scalability for OAI-PMH load balancing community building discovery

55. OAI-PMH data model resource item Dublin Core metadata MARCXML metadata records entry point to all records pertaining to the resource metadata pertaining to the resource OAI-PMHidentifier metadataPrefix datestamp OAI-PMH identifierOAI-PMH sets

56. OAI-PMH Used by Google & AcademicLive (MSN) Why support OAI-PMH? $These guys are in business (i.e., for profit) QHow does OAI-PMH help their bottom line? ABy improving the search and analysis process

57. Resource Harvesting with OAI-PMH resource item Dublin Core metadata METS records OAI-PMH identifier = entry point to all records pertaining to the resource MPEG-21 DIDL metadata pertaining to the resource simplehighly expressive more expressive highly expressive MARCXML metadata

58. OAI-PMH mod_oai: complex objects + resource harvesting Research Focus Outline: Web Server Enhanced Preservation

59. Two Problems The counting problem There is no way to determine the list of valid URLs at a web site The representation problem Machine-readable formats and human- readable formats have different requirements

60. Integrate OAI-PMH functionality into the web server itself… mod_oai: an Apache 2.0 module to automatically answer OAI-PMH requests for an http server –written in C –respects values in.htaccess, httpd.conf compile mod_oai on http://www.foo.edu/http://www.foo.edu/ baseURL is now http://www.foo.edu/modoai –Result: web harvesting with OAI-PMH semantics (e.g., from, until, sets) mod_oai solution The human- readable web site Prepped for machine-friendly harvesting http://www.foo.edu/modoai?verb=ListIdentifiers&metdataPrefix=oai_dc&from=2004-09-15&set=mime:video:mpeg Give me a list of all resources, include Dublin Core metadata, dating from 9/15/2004 through today, and that are MIME type video-MPEG.

61. A Crawler’s View of the Web Site Not crawled (unadvertised & unlinked) web root Crawled pages Not crawled (too deep) Not crawled (protected) Not crawled (remote link only) Not crawled (Generated on-the-fly by CGI, e.g.) Not crawled robots.txt or robots META tag Remote web site

62. Apache’s View of the Web Site web root Require authentication Unknown/not visible Generated on-the-fly (CGI, e.g.) Tagged: No robots

63. The Problem: Defining The “Whole Site” For a given server, there are a set of URLs, U, and a set of files F –Apache maps U  F –mod_oai maps F  U Neither function is 1-1 nor onto –We can easily check if a single u maps to F, but given F we cannot (easily) generate U Short-term issues: –dynamic files exporting unprocessed server-side files would be a security hole –IndexIgnore httpd will “hide” valid URLs –File permissions httpd will advertise files it cannot read Long-term issues –Alias, Location files can be covered up by the httpd –UserDir interactions between the httpd and the filesystem

64. Tagged: No robots A Webmaster’s Omniscient View web root Deep Dynamic Authenticated Orphaned Unknown/not visible MySQL 1.Data1 2.User.abc 3.Fred.foo httpd 1.file1 2./dir/wwx 3.Foo.html

65. Machine-readable Human-readable HTTP “Get” versus OAI-PMH GetRecord mod_oai HTTP GET HTTP GetRecord JHOVE METADATA MD-5 LS Complex Object WEB SITE Apache Web Server “GET /headlines.html HTTP1.1” “GET /modoai/?verb=GetRecord&identifier= headlines.html&metadaprefix=oai_didl”

66. OAI-PMH data model in mod_oai resource item Dublin Core metadata records OAI-PMH identifier = entry point to all records pertaining to the resource MPEG-21 DIDL metadata pertaining to the resource HTTP header metadata http://techreports.larc.nasa.gov/ltrs/PDF/2004/aiaa/NASA-aiaa-2004-0015.pdf OAI-PMH sets MIME type

67. Complex Objects That Tell A Story First came Lenin Then came Stalin… Resource and metadata packaged together as a complex digital object represented via XML wrapper Uniform solution for simple & compound objects Unambiguous expression of locator of datastream Disambiguation between locators & identifiers OAI-PMH datestamp changes whenever the resource (datastreams & secondary information) changes OAI-PMH semantics apply: “about” containers, set membership Russian Nesting Doll http://foo.edu/bar.pdf encoded as an MPEG-21 DIDL............... SADLFJSALDJF...SLDKFJASLDJ Jhove metadata DC metadata Checksum … Provenance

68. Resource Discovery: ListIdentifiers HARVESTER: issues a ListIdentifiers, finds URLs of updated resources does HTTP GETs updates only can get URLs of resources with specified MIME types

69. Preservation: ListRecords HARVESTER: issues a ListRecords, Gets updates as MPEG- 21 DIDL documents (HTTP headers, resource By Value or By Reference) can get resources with specified MIME types

70. What does this mean? For an entire web site, we can: –serialize everything as an XML stream –extract it using off-the-shelf OAI-PMH harvesters –efficiently discover updates & additions For each URL, we can: –create “preservation ready” version with configurable {descriptive|technical|structural} metadata e.g., Jhove output, datestamps, signatures, provenance, automatically generated summary, etc. Harvest the resource extract metadata include an index translations… or lexical signatures, Summaries, etc Jhove & other pertinent info Wrap it all together In an XML Stream Ready for the future

71. OAI-PMH mod_oai: complex objects + resource harvesting Research Focus Outline: Web Server Enhanced Preservation

72. Research Contributions Thesis Question: How well can Apache support web page preservation? Goal: To make web resources “preservation ready” –Support refreshing (“how many URLs at this site?”): the counting problem –Support migration (“what is this object?”): the representation problem How: Using OAI-PMH resource harvesting –Aggregate forensic metadata Automate extraction –Encapsulate into an object XML stream of information –Maximize preservation opportunity Bring DL technology into the realm of WWW

73. Experimentation & Evaluation Research solutions to the counting problem –Different tools yield different results –Google sitemap <> Apache file list <> robot crawled pages –Combine approaches for one automated, full URL listing Apache logs are detailed history of site activity Compare user page requests with crawlers’ requests Compare crawled pages with actual site tree Continue research on the representation problem –Integrate utilities into mod_oai (Jhove, etc.) –Automate metadata extraction & encapsulation Serialize and reconstitute –complete back-up of site & reconstitution through XML stream

74. Summary: Web Server Enhanced Preservation Better web harvesting can be achieved through: –OAI-PMH: structured access to updates –Complex object formats: modeled representation of digital objects Address 2 key problems: –Preservation (ListRecords) – The Representation Problem –Web crawling (ListIdentifiers) – The Counting Problem mod_oai: reference implementation –Better performance than wget & crawlers –not a replacement for DSpace, Fedora, eprints.org, etc. More info: –http://www.modoai.org/http://www.modoai.org/ –http://whiskey.cs.odu.edu/http://whiskey.cs.odu.edu/ Automatic harvesting of web resources rich in metadata packaged for the future Today: manualTomorrow: automatic!

75. Summary Michael L. Nelson

76. Summary Digital preservation is not hard, its just big. –Save the women and children first, of course, but there is room for many more… Using the by-product of SE and WI, we can get a good amount of preservation for free –prediction: Google et al. will eventually see preservation as a business opportunity Increasing the role of the web server will solve most of the digital preservation problems –complex objects + OAI-PMH = digital preservation solution

77. “As you know, you preserve the files you have. They’re not the files you might want or wish to have at a later time” “if you think about it, you can have all the metadata in the world on a file and a file can be blown up” image from: http://www.washingtonpost.com/wp-dyn/articles/A132-2004Dec14.html

78. Overview of OAI-PMH Verbs VerbFunction Identifydescription of repository ListMetadataFormatsmetadata formats supported by repository ListSetssets defined by repository ListIdentifiersOAI unique ids contained in repository ListRecordslisting of N records GetRecordlisting of a single record metadata about the repository harvesting verbs most verbs take arguments: dates, sets, ids, metadata formats and resumption token (for flow control)

79. Enhancing Apache’s utility as a preservation tool Create a partnership between server and SE –Apache can serve up details about site, accessible portions of site tree, changes including additions and deletions –SE can reduce crawl time and subsequent index/update times Google: “Hi Apache! What’s new?” Apache: “Hi Google! I’ve got 3 new pages, xyz/blah1.html, yyy/bug2.html, and ru2.html. Oh, and I also deleted xyz/boo.html.” Use OAI-PMH to facilitate conversation between the SE and the server –Data model offers many advantages Both content-rich and metadata-rich Supports complex objects – Protocol’s 6 verbs mesh well with SE, Server roles ListMetadataFormats, ListSets, GetRecord, ListRecords, ListIdentifiers, ListRecords Enable policy-driven relationship between site & SE –push content-rich harvesting to web community

80. OAI-PMH Entityvaluedescription ResourceURLPDF, PS, XML, HTML or other file Item identifierOAI Identifier DNS-based name of metadata about resource set membershipLCSHLibrary of Congress Subject Heading Record metadataPrefixoai_dcbibliographic metadata in Dublin Core datestamp2004-10-18modification date of DC record Record metadataPrefixoai_marcbibliographic metadata in MARC datestamp2004-07-31modification date of MARC record OAI-PMH concepts : typical repository

81. OAI-PMH Entityvaluedescription ResourceURLHTML, GIF, PDF or other web file Item identifierURLsame URL as the resource set membershipMIME typeMIME type of the resource Record metadataPrefixhttp_headerthe http headers that would have been returned via HTTP GET/HEAD datestamp2004-07-31modification date of resource Record metadataPrefixoai_dca subset of http_header in DC datestamp2004-07-31modification date of resource Record metadataPrefixoai_didlMPEG-21 DIDL: base64 encoded resource + http_header metadata datestamp2004-07-31modification date of resource OAI-PMH concepts : mod_oai

82. OAI-PMH data model resource item Dublin Core metadata METS records OAI-PMH identifier = entry point to all records pertaining to the resource MPEG-21 DIDL metadata pertaining to the resource simplehighly expressive more expressive highly expressive MARCXML metadata

83. Warrick API API should provide a clear and flexible interface for web repositories Goals: –Shield Warrick from changes to WI –Facilitate inclusion of new web repositories –Minimize implementation and maintenance costs

84. Evaluation Internet Archive has endorsed use of Warrick Make Warrick available on SourceForge Measure the community adoption & modification

85. performance of mod_oai and wget on www.cs.odu.eduwww.cs.odu.edu for more detail: “mod_oai: An Apache Module for Metadata Harvesting “ http://arxiv.org/abs/cs.DL/0503069

86. IndexIgnore & File Permissions

87. Alias: Covering Up Files httpd.conf: Alias /A /usr/local/web/htdocs/B Alias /B /usr/local/web/htdocs/A the files “A” and “B” will be different from the URLs http://server/A http://server/B

88. UserDir: “Just in Time” mounting of directories whiskey.cs.odu.edu:/ftp/WWW/conf% ls /home liu_x/ mln/ whiskey.cs.odu.edu:/ftp/WWW/conf% ls -d /home/tharriso /home/tharriso/ whiskey.cs.odu.edu:/ftp/WWW/conf % ls /home liu_x/ mln/ tharriso/ whiskey.cs.odu.edu:/ftp/WWW/conf %

89. Complex Object Formats : Characteristics Representation of a digital object by means of a wrapper XML document. Represented resource can be: –simple digital object (consisting of a single datastream) –compound digital object (consisting of multiple datastreams) Unambiguous approach to convey identifiers of the digital object and its constituent datastreams. Include datastream: –By-Value: embedding of base64-encoded datastream –By-Reference: embedding network location of the datastream –not mutually exclusive; equivalent Include a variety of secondary information –By-Value –By-Reference –Descriptive metadata, rights information, technical metadata, …

90. Resource Harvesting: Use cases Discovery: use content itself in the creation of services –search engines that make full-text searchable –citation indexing systems that extract references from the full-text content –browsing interfaces that include thumbnail versions of high-quality images from cultural heritage collections Preservation: –periodically transfer digital content from a data repository to one or more trusted digital repositories –trusted digital repositories need a mechanism to automatically synchronize with the originating data repository Ideas first presented in Van de Sompel, Nelson, Lagoze & Warner, http://www.dlib.org/dlib/december04/vandesompel/12vandesompel.html