Web caching is not a burning issue in IT. A few people are motivated to improve caching techniques for the sake of network efficiency. But the overwhelming majority of IT professionals would sooner have their gums scraped than review the literature on Web caching. Like such other arcane areas in networking (load balancing, TCP/IP protocols, Domain Name Servers, and network topologies come to mind), most of us are content to know that these issues are very important to our communication practices, and that very bright, industrious people are looking after our interests by studying them. Web caching, however, falls into another category altogether. Like cookies, Web Caching protocols are examples of how the best of intentions can result in implementations with highly undesirable side-effects. One aspect of modern Web caching is just socially irresponsible, pure and simple.
In the world of networks, caching is an important optimization tool. It shares this trait with replication and mirroring. Networks generate a lot of background "chatter" to keep the data moving along smoothly, and if it weren't for these optimization tools the volume of chatter would overwhelm the useful data, and communication would come to a halt.
DNS performance is a good example. When we access a computer on the Internet, we are actually targeting a unique computer with a unique 32-bit IP address broken out into 4 octets. Thus, "how.what.why.when.where.com" would be resolved to an address like "188.8.131.52." This name resolution takes place in a Domain Name Server -- more accurately, that translation takes place somewhere within a hierarchy of Domain Name Servers that collectively make up a large distributed database. DNS was set up as a distributed database because the pioneers of Internetworking discovered in the early days that managing DNS names was an inherently non-scalable activity.
So, when we attempt a connection with "how.what.why.when.where.com" we must rely on authoritative address resolution into IP addresses or address fragments (domains). But, there will likely be several DNS servers that have authority over the respective domain fragments involved: one authoritative root server for the top level domain (.com), another DNS server for the next lower domain (.where), and so forth. This model will work fine as long as these domains are linked together within the hierarchy, so that each DNS server knows how to reach authoritative name servers for the parent and child nodes. In our example, the complete address resolution of "how.what.why.when.where.com" would entail the iterative process of getting the address for the authoritative DNS server for "where.com" from the ".com" root server, then getting the authoritative DNS server for "when.where.com" from the DNS server for "where.com" and so forth down the tree. However, these DNS servers may be scattered all over the Internet. Therein lies the rub. Without optimization, our authoritative root server for ".com" is going to receive a request every time someone tried to access any computer in that domain - it couldn't handle all this traffic.
Enter caching. Caching shares "bindings" of computer names with their corresponding IP addresses after these have been resolved with other DNS servers that have accessed the information? When I access "how.why.what.when.where.com," the local authoritative name server stores the binding of "how.what.why.when.where.com" with "184.108.40.206" in its local cache. The principle of locality of references suggests that if I access this computer once, I'm likely to do it again, so caching the binding will unnecessary address resolution traffic off the network above the level of the local server. I just have to add an instruction to my network OS to look to the local cache for address resolution before passing on the request up the hierarchy of DNS servers.
DNS caching falls into the category of benign caching. Similar benign caching takes place at the lower, physical network layer through the Address Resolution Protocol (ARP), only there the bindings are from IP addresses to actual physical addresses of the computer built into the hardware of the motherboard or network card (type ipconfig /all> at a command prompt to find yours).
Web caching goes beyond the benign. In its extreme form, it routinely violates the principles that modern societies routinely use to manage intellectual property up to and including overt copyright infringement. For want of a better term, we'll refer to this as the malignant strain of Web caching.
This is not to deny that there are benign aspects of Web caching. In its most benign form, Web caching achieves efficiency through a document delivery environment built upon content and location transparency. Other things equal, if we can get current versions of sought-after information more efficiently, what's not to like? Of course the hook is "other things equal." More on that in a moment.
Two examples of increased efficiency of Web document access that satisfies the simultaneous objectives of increased efficiency and content and location transparency are server caches and browser caches. In the first case, servers may cache frequently accessed documents in primary memory to avoid slow disk accesses, or server farms may place high-demand data on faster, dedicated servers for load balancing.
In the second case, the documents are served actually served on the client, itself. An example of this is a browser cache. The browser stores frequently accessed data on the client's hard disk so that subsequent access can entirely avoid network access altogether. As with server caching there is no issue with subsequent reuse or re-distribution of the resource or document that is inconsistent with the owner/author's objectives. The client's browser cache behaves in much the same way that a local DNS server would use name resolution bindings. Server-caching and browser caching are both (a) ephemeral, and (b) directly linked to the original source.
Such is not the case with the potentially pernicious form of Web caching on proxy servers. Here matters become murky. Proxy servers are network "middlemen" that stand between computers. Typically, a proxy server would be located between a client and server. In such case, the proxy would interact with the server on the client's behalf, and the other way around. Besides proxy caches, other examples of proxy servers are proxy firewalls, "anonymizers," and remailers.
Proxy servers simultaneously satisfy the two conditions of efficiency and content and location transparency, but are neither ephemeral nor directly linked to the original source. The proxied cache becomes a surrogate for the original source. This surrogate does not come without ethical implications.
The original version of Web supported in HTTP 1.0 was fairly anemic. While the implementation details are only of historical interest at this point, suffice it to say that the three relevant headers were included to either (a) determine whether requested documents were "fresh" or "stale" with the end of circumventing the cache in favor of the originating host if the documents in question were out-of-date, or (b) prevent documents from being cached at all.
HTTP 1.1's treatment of Web caching is much more sophisticated. Considerable flexibility has been added for making finer-grained distinctions between "fresh" and "stale" documents, including the use of heuristics rather than server-supplied file time-and-date stamp comparisons. A validation model was included that added a "validator token" to the cached document. With a cache-resident validator, document currency could be confirmed by the proxy cache by simply running the validator by the originating server. The currency is reported by the response status code. In fact, there are two types of validators, and correspondingly, two degrees of validation: strong and weak, depending upon the degree of "freshness" one needs. But by far the most important difference in the HTTP 1.1 treatment of caches was the addition of the cache-control header field.
The general idea between the cache-control header field was to provide a general mechanism for caches to communicate efficiently and effectively with other computers. The point that should not be overlooked is that by design the cache-control header field provides directives that are focused on regulating requests and responses along the caching food chain. The general structure of the header is:
Cache-control : directive [optional parameters]
Where the directive consists of keywords relating to aspects of requests and responses. Since the more controversial aspect of cache-control deals with responses, we'll limit our discussion accordingly:
Cache-control = "public" meaning: responses from this server may be
stored without restriction
Cache-control = "max-age = 43,200" meaning: the document contained in this response should be considered "stale" after 43,200 seconds (approx. 30 days)
Cache-control = "must re-validate" meaning: the caching service (proxy or browser) must re-validate the document after it becomes "stale" from the originating server, or report an error message. It must not pass on the "stale" version.
You get the idea. Now, the worrisome directives:
Cache-control = "private" meaning: only private cache services (not
shared cache services) may cache the contents of this
Cache-control = "no-cache" meaning: responses can be cached , but the cached copy may not be reused for subsequent requests without re-validating the cached copy with the originating server (n.b., the baroque logic behind this directive is a work of art. When we say "don't peek" does that normally mean "peek only once"?)
Cache-control = "no-store" meaning: this response may not be cached on non-volatile storage.
It doesn't take a rocket scientist to review the list of Internet Cache Protocol cache-control directives and determine that its approach to intellectual property management falls somewhere between irresponsible and draconian. Proponents of Web caching in its present form subscribe to the "anything on the Web is community property" school of thought. This is made abundantly clear in the crude way that document ownership is skirted in the last three directives discussed above.
Let me suggest some topics that are directly relevant to the issue of ownership, together with what many of us consider reasonable concomitant rights and responsibilities, for which there are no provisions in the Internet Cache Protocol:
1. Does the cache operator own the resource/document being accessed?
2. Is this resource/document copyrighted, and if so what is the nature of the license for subsequent re-distribution..
3. Is there a "definitive" version of this document (e.g., the published and copyrighted version)? If so, where is it located?
4. Does the cached document have a Digital Object Identifier? If so, what is it?
5. Did the owner/copyright holder restrict the use of this resource/document (e.g., for classroom use, for use by not-for-profit corporations, for use other than commercial purposes)?
6. Is it realistic to expect the owner/author to familiarize him/herself with the nuances of the cache-control directives before placing anything on the Web?
7. Should the owner/author be expected to define the caching parameters that take into account all possible effects (e.g., Would the typical author interpret "no-cache" to mean "don't cache without re-validating"?)
8. Did the owner/author relinquish control over when a resource/document may be withdrawn from public circulation by placing it on the Web? (n.b., once the document is cached, there's no simple and immediate way of withdrawing all cached copies from the Web.)
9. Is the owner/author entitled to maintain accurate usage statistics and logs regarding resources of their creation? (These statistics are more difficult to gather from caches, demand additional work, may (as in the cost of embedded Web bugs) have privacy implications, carry with them a definite cost to the owner/author/information provider, and require the universal cooperation of cache maintainers to work. Cache metering is not catching on for these reasons.)
10. Is it the responsibility of the owner/author to ensure that version control mechanisms are in place that prevent outdated versions from being circulated by cache services?
11. Is it the responsibility of the owner/author to continuously monitor every proxy cache operation and issue a "takedown" notice each time some of their intellectual property is discovered in some cache?
12. Is the owner/author of a Web document/resource entitled to reasonable royalties or profit sharing that accrue to downstream Web caching services that host their work?
I predict that answers to these sorts of questions will determine how sympathetic one is to the objectives of proxy cache services. Regrettably, very few people seem to even ask these questions, much less try to develop viable answers.
Those of us who oppose Web caching in its present form run the risk of being labeled "cache busters." But the label just doesn't fit. I know of no one who opposes the technology of caching as such and in general. But, like cookies, caching technology is being misused. That fraction of the Web community that respects the ownership of intellectual property may be forced into cache busting because there are no other reasonable alternatives in the cache-control directives to protect their intellectual property. It is worth remembering that under HTTP 1.1, "no-store" is the only option available for those who want to control the dissemination of their intellectual property. No-store is far too coarse to be useful to the Web community in general.
The no-store directive shares this deficiency with the robot exclusion standard that was built on the premise that the primary objective of agent indexing controls should be the convenience of the system administrators. The idea that the millions of Web developers would organize their sites on the basis of what should be indexed by search agents rather than natural semantic clusters is preposterous. But that's the only way that the robot exclusion standard can work. So it is with the binary "no store" directive. Web developers are given the choice of either preventing any caching or giving up complete control of their intellectual property. I'm sorry, but the real world doesn't limit us to a succession of equally unpleasant choices in a way the "no-store" directive does.
The defenders of intellectual property rights face formidable foes. The Web cache community seems to embrace the position that there should be no expectation of proprietary ownership for anything placed on the Web. Usually this takes the form of remarks like "the Web is about copying - if you don't want your work copied, don't put it on the Web." This is an exceeding simplistic and narrow-minded view of the Web.
While the world recognizes a difference between perusing the intellectual property of others on the one hand, and making and distributing copies of it on the other, this distinction seems to have completely escaped the attention of the creators of the Internet Cache Protocol. When a teacher provides their students with perusal copies of written work, designs, computer programs, artwork, etc., there is no implication that they surrender any ownership rights. Nor is there any implication that the teacher has given the students a license to copy, distribute or sell copies. To claim that there is such an implication would be patently absurd. Artists don't automatically surrender rights when they agree to have their paintings displayed in a gallery. Neither does the owner/author of Web content.
Google is a clear example of this abuse. One of the things that made Google distinct from search engines was that they cached the content they indexed - without, mind you, any regard to the questions I posed in the previous section. Google generates revenue from having a very large index of cached intellectual property that they don't own and, for the most part, haven't received permission to cache and re-distribute. Some of us have a problem with this: publishers in particular. In one of life's little ironies, the publishers and authors are not necessarily on the same page on this issue.
To illustrate, I'll use the example of publishing an article with ACM. Let's think of a very linear model of producing a publication (for description of the more realistic, non-linear model see my "Cyberpublishing Manifesto" in the March, 2001 issue of CACM). On this account, an author completes some research or comes up with a new idea and commits it to writing. Continued reflection and input from external sources produces k iterations of this document, only the last of which is submitted to ACM for editorial consideration. ACM sends the material out to reviewers, adds editorial value, requests revisions from the author, formally accepts the k+1st version, copyrights it, assigns a unique document object identifier (DOI), and publishes it as the "definitive" version. Along the way, a subset of the first k versions are likely to have found their way onto the Web.
The ACM copyright policy (by anyone's measure a very author-friendly policy) is clear on several critical points relating to this process. For one, if any of the first k versions are substantially the same as the accepted version, the ACM copyright notice and a link to the definitive, copyrighted version must be attached after the article is accepted. Google doesn't recognize this copyright arrangement at all. In fact, under the current contractual arrangement with ACM, Google only agrees not to serve up "full text" versions of ACM "definitive" copies. There is no mention at all of prior versions that bear the ACM imprimatur, never mind any consideration given to the earlier versions that may be copyrighted by the author. Under the current Web caching scheme, the author has no protection against copyright infringement at all.
The Devil's in The Detail
The optimization side of Web caching is indispensable to the successful deployment of the Internet. The intellectual property side is a disaster.
The solution to the problem is the engagement of people who are sensitive to intellectual property issues, not just those interested in improving network efficiency, in a complete overhaul of the Internet Caching Protocol. The new version should explicitly contain Cache-control directives that are socially responsible when it comes to authorship, providing appropriate credit, delineating just how much latitude the owner/author has licensed regarding subsequent distribution, whether royalties are required, definitive versions, copyright, and so forth.
As a modest first step, consider directives like these:
Credits Required = list
Copyrighted = "yes/no"
Copyright holder = name
DOI = value
Serveup = "full text | extract | keywords | URL" [inclusive or]
Extract criteria = "word limit = value", "document abstract = yes/no", etc.
Royalty required = "yes/no"
Methods of royalty payment = list
The idea is to put the control over the distribution of the intellectual content of the Web in the hands of those who created it in the first place.
Generic information about Web caching, including recent news flashes, may be found on the Internet Caching Resource Center Website at http://www.caching.com/.
Internet Cache Protocol specifications may be found at icp.ircache.net.
The International Web Content Caching and Distribution Workshops have been held annually since 1996. Program information and proceedings are online at http://www.iwcw.org/. The focus of these workshops is technical, with no evidence that I can find that there is much attention paid to underlying ethical issues.
Duane Wessels Web Cache site is online at www.web-cache.com. Wessels penned what is to my knowledge the most complete reference book on Web Caching for O'Reilly in 2001. He also co-authored version 2 of the RFC for Internet Caching Protocol specifications. Wessels pays lip service in both his book and Website to the ethical and privacy issues of Web caching.
Many caching programs for a wide variety of Unix and Windows NT/2000/XP servers are available. These include:
The Open Source Squid Web Proxy Cache (http://www.squid-cache.org/)
Netscape's Proxy Server (wp.netscape.com/proxy/v3.5/)
Microsoft's caching software is built into their Internet Security and Acceleration Server product (see www.microsoft.com/isaserver).
The CiscoCache Engine Series: (http://www.cisco.com/warp/public/cc/pd/cxsr/500/index.shtml)
The ACM copyright policy is online at http://www.acm.org/pubs/copyright_policy/. Digital Object Identifiers are described at http://www.doi.org/.