Re: Expiration scans...

Duane Wessels wrote:
>
>cb@brewhq.swb.de writes:
>
[what is...]
>>---
>>96/10/29 07:25:43| Completed 3 full expiration scans of store table
>>---
>It scans the entire cache roughly every 24 hours, one bucket at a time.
>It removes expired objects, ie:
>
> negative-cached objects
>
How would these wind up there anyway? From looking at the beta11 store.log
objects have either a 0 or -1 time assigned to them...

> objects which have not been referenced for longer than 'reference_age'
>
>The 'reference_age' parameter is disabled by default, and alas undocumented.
>
>The idea is to have reference_age be automatically adjusted to keep
>the cache size between high/low water marks and avoid the need to enter
>the emergency-object-deletion mode.
>
This is a nice goal, not having execeeded the high water mark yet I wonder
how much of a punishment this emergency-object-deletion mode would be.
Basically I want to keep data as long as possible, since an IMS request
is always preferable to a reload of all the data...

Oh, something odd with beta11 and certain pages, if the page has no
LM info (like http://altavista.digital.com/ ) _all_ objects on it are
subject to an IMS check as well, even though they're quite well within
my specified freshness parameters. This introduces unwanted and uncessary
latency of course. I found one page where this also happens with a LM
date of last month, gotta see what's different about this...

I'd like to urge again all proxy maintainers to notify the webmasters of
sites which do not supply Last Modified data, doing so will hopefully
raise awareness for this topic and make our lives easier. ^_^

Mata ne,

<CB>

-- 
  // <CB> aka Christian Balzer, Tannenstr. 23c, D-64342 Seeheim, Germany
\X/  CB@brewhq.swb.de | Voice: +49 6257 83036, Fax/Data: +49 6257 83037
SWB  - The Software Brewery - | Team H, Germany HQ | Anime no Otaku