- From: Jeffrey Mogul <mogul@pa.dec.com>
- Date: Thu, 21 Dec 95 15:37:35 PST
- To: http-caching@pa.dec.com
Since I'm about to disappear for 6 days or so, I thought I should send this out. I put it together with some help from several of you, but it's probably fairly cryptic at this point and not complete. Suggested additions are welcome. -Jeff Basic caching model issues: What is the unit of caching? (e.g., "entity + response headers") What is the cache key? How flexible is this? How does the cache determine if an entry is valid? When must/should validation be done? What does Expires: do? What does Cache-control: do? What happens if validation isn't possible? How can UAs override default policies? Compatibility with 1.0 implementations (If-Modified-Since) How is caching supposed to work with content negotiation? Can the cache tell the server something that will result in the use of a cached item instead of a variant? Security issues Spoofing using Location headers (prevention thereof?) [or leave this to the content-negotiation subgroup?] Privacy issues TBS Compatibility with post-1.0 implementations Is anyone using Cache-control? If so, what syntax+semantics? Which response-headers are kept in the cache? who gets to decide? is this a priori (i.e., listed in the spec) or dynamic? Review "proxy authentication" (requested by David Morris) Multiple URLs for the same object: should the cache avoid storing multiple copies? Should the protocol make copy-combination easier? Are there invalidation problems? Extension methods: Can we define a reliable and secure mechanism that allows proxies to forward methods that they do not understand, and perhaps to cache the results? Cache-prevention: Can providers ensure that no copy of an object is cached anywhere? This is for content-providers who are worried about leakage of sensitive information. History-prevention: can the origin-server control what goes into the history list on a browser? (More precisely, can/should a response specify "not to be placed in history list"?) [Maybe not a topic for this subgroup] Cached-copy-update problem: How to preserve cache consistency when an update is made. For example, what happens if an obejct is cached in two different proxies and one client does an update? Tradeoff between validation traffic and stateness: given that this is unavoidable: how can the protocol design optimize things? who gets to decide how to set the tradeoff point? how does the protocol allow the tradeoff point to be set?
Received on Thursday, 21 December 1995 23:44:53 UTC