[
https://issues.apache.org/jira/browse/TS-2020?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13704665#comment-13704665
]
Phil Sorber commented on TS-2020:
---------------------------------
To elaborate a little, right now there are two ways something gets evacuated
(re-written behind the write pointer). 1) Actively being read and delivered to
a client, or 2) explicitly "pinned" in the cache. I would like to add a third
method, to see if it is in the ram cache, and if it is we assume it's popular
and worth keeping in the cache and not overwriting. Sort of a poor mans LRU. As
long as your ram cache is not close to the same size as your disk cache it
should not thrash.
Also, from IRC:
09:24 <zwoop1> we were going to start brainstorming on the v3.6 "roadmap"
09:26 <@PSUdaemon> what about adding ram cache lookup for disk cache evacuation
heuristic?
09:27 <@PSUdaemon> to get psuedo LRU/CLFUS behavior in the disk cache
09:27 <zwoop1> PSUdaemon: yeah, I like that (and john approves too)
> Add check to evacuation code to do RAM cache lookup and evacuate on a hit
> -------------------------------------------------------------------------
>
> Key: TS-2020
> URL: https://issues.apache.org/jira/browse/TS-2020
> Project: Traffic Server
> Issue Type: Improvement
> Reporter: Phil Sorber
> Labels: C
> Fix For: 3.5.0
>
>
> This will allow us to keep hotter items in the disk cache longer. It's not a
> full LRU but it's cheap and will help.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
