[
https://issues.apache.org/jira/browse/TS-5090?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15745966#comment-15745966
]
Leif Hedstrom commented on TS-5090:
-----------------------------------
I attached a graph, showing how the CFLUS from server restarts deteriorates
over time vs the LRU RAM cache. We observe a similar change in disk I/O
utilization, where the I/O increases by a factor of 3x with the CLFUS.
> Cache hit ratio deteriorating over time with CLFUS
> --------------------------------------------------
>
> Key: TS-5090
> URL: https://issues.apache.org/jira/browse/TS-5090
> Project: Traffic Server
> Issue Type: Bug
> Components: Cache
> Reporter: Leif Hedstrom
> Attachments: RAM Cache.png
>
>
> We've observed poor cache hit ratio with the CLFUS, where it starts doing
> great, but over ~1 day, the cache hit ratio deteriorates. It's to the point
> where the LRU + seen-filter=1 is 2-3x better. In addition, the RAM cache hit
> ratio on the LRU stays consistent over time.
> At first, I thought it was only our particular use case, but it seems others
> have confirmed the same behavior.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
