- From: Patrick McManus <mcmanus@appliedtheory.com>
- Date: Fri, 22 Nov 1996 09:08:54 -0500 (EST)
- To: Jeffrey Mogul <mogul@pa.dec.com>, http working group <http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com>
In a previous episode...Jeffrey Mogul said: -> -> > A proxy SHOULD NOT transmit "Meter: count=0/0", since this conveys -> > no useful information. -> The origin server is explicitly given the right the right to decide -> whether or not to enter into a caching 'contract' with a proxy based -> upon any criteria it chooses. I can envision a situation where -> reporting of 0 hit counts would be a valuable criteria to base this -> decision on. -> -> You may be right that a case could be made that this information -> is sometimes useful. But if so, then I think you need to suggest -> a modification to the proposal that would allow the server to -> request the 0/0 counts *only* when they are needed. And this would -> have to include some careful language around how and when a proxy -> receiving such a report needs to forward it to the next inbound -> server. -> -> In the absence of a specific server request for a 0/0 count, -> the proxy would not know if the 0/0 report is worth sending. That's absolutely true, and I think you can take that conclusion one step further. In the absence of a specific server request for any minimum return count, the proxy cannot know if the report is worth sending. Some servers may not believe that a return connection is worth the overhead of receiving a 1/0 report, just as some servers may insist on seeing a 0/0 report to know that the proxy's 'best-effort' is succeeding. perhaps the initial negotiation could contain a count-range meter header from the proxy indicating the bounds (both hi and low) it is willing to report for, and the server response could return min and max values in that range that represents the reporting interval. (i.e. don't report anything less than 2 hits, but never go more than 15 without sending a report) This nicely separates the functionality of max-uses from the reporting functions in any case. I don't see any reason to believe that their values are going to be derived from the same criteria, just because you want more frequent reports from a cache that has held your page for 2 weeks than every 2 weeks, doesn't mean the resource is now invalid. What do people think of this? -> I think there are better metrics for measuring network reliability. -> For example, with just a little help from the TCP stack, either -> end could count the number of retransmissions and/or timeouts. agreed, but my point is that the server's decision to enter into the reporting 'contract' and allow caching is a decision that is opaque to your proposal. There shouldn't be language that excludes some techniques because it isn't relevant or important, in a black-boxed decision we can't know what will be relevant. -Patrick -- Patrick R. McManus - Applied Theory Communications - Software Engineering http://pat.appliedtheory.com/~mcmanus Programmer Analyst mcmanus@AppliedTheory.com 'Prince of Pollywood' Standards, today! *** - You Kill Nostalgia, Xenophobic Fears. It's Now or Neverland. - ***
Received on Friday, 22 November 1996 06:20:48 UTC