cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Benedict (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-11452) Cache implementation using LIRS eviction for in-process page cache
Date Fri, 15 Apr 2016 23:32:25 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-11452?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15243842#comment-15243842
] 

Benedict commented on CASSANDRA-11452:
--------------------------------------

Yeah, another possibility is to pick a random hash table bucket's first element for the guard,
or do this for a proportion of admissions.

Some overly complex options are also to maintain a mean/median of the last X LRU items, or
to decide a %ile given the sketch and map size, calculate it from the sketch periodically,
and admit only those above it.

However I'm not sure Cassandra cares too much about a concerted attacker for this since it
would be very hard to predict what the keys would be (you'd need the random seed, the existing
sstable layouts, other keys and the ability to directly induce load on the server, and even
then it would be challenging).  So long as there is no realistic accidental way for it to
happen I'd say we're good.

One slight variant to consider is using an xor of the victim and candidate hashes for our
RNF.  This is likely quicker, should be perfectly safe to attack for any good hash function
(especially post-spread), and would have the nice property of determinism for unit tests.

> Cache implementation using LIRS eviction for in-process page cache
> ------------------------------------------------------------------
>
>                 Key: CASSANDRA-11452
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-11452
>             Project: Cassandra
>          Issue Type: Improvement
>          Components: Local Write-Read Paths
>            Reporter: Branimir Lambov
>            Assignee: Branimir Lambov
>
> Following up from CASSANDRA-5863, to make best use of caching and to avoid having to
explicitly marking compaction accesses as non-cacheable, we need a cache implementation that
uses an eviction algorithm that can better handle non-recurring accesses.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message