[
https://issues.apache.org/jira/browse/TS-3174?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14203330#comment-14203330
]
Zhao Yongming commented on TS-3174:
-----------------------------------
hmm, are you sure you get the correct understanding of the CLFUS effects? In
our using of the ram cache, the CLFUS will cause some trouble on memory
wasting, especially with a heavy changed traffic patten, be cause that CLFUS
will try to cache more small objects and swapping out the big objects after ram
cache memory full. that is a good feature, but need more working to archive the
memory allocating/de-allocating during this step, I think that there still need
more work to make the big objects swaping out and de-allocate or reuse.
I know that will not kill TS on most so busy systems, but you still need to
keep an eye on that. TS cop process will bring back the failed server, it may
hide most of the problems from users. :D
and it is easy to verify, on system with mixed objects, ie, active object size
range from 1KB-100MB.
set a higher ram cut off size from 4M to 100M, and following the
doc/sdk/troubleshooting-tips/debugging-memory-leaks.en.rst to enable memory
dump, compare the allocated and used memories on each size.
FYI
> Kill LRU Ram Cache
> ------------------
>
> Key: TS-3174
> URL: https://issues.apache.org/jira/browse/TS-3174
> Project: Traffic Server
> Issue Type: Task
> Reporter: Susan Hinrichs
> Fix For: 6.0.0
>
>
> Comment from [~zwoop]. Now that CLFUS is both stable, and default, is there
> even a reason to keep the old LRU cache. If no objections should remove for
> the next major version change.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
