[
https://issues.apache.org/jira/browse/CASSANDRA-8897?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14487023#comment-14487023
]
Benedict commented on CASSANDRA-8897:
-------------------------------------
bq. If we adopted increasing slab sizes how do we know when it's time to
increase the size
Each time we exhaust the current slab. i.e. we allocate a 64K slab first, then
a 128K, 256K, 512K, and then we stick with 512K/1Mb from then on. That said, if
we impose a maximum buffer size of 64Kb, we could stick to a fixed 64Kb "slab"
that is simply sliced from a larger slab. i.e. Our shared pool allocates slabs
of 1Mb, we slice it into 64Kb chunks, and share those chunks, thus amortizing
the page alignment costs. For simplicity we want to ensure that we have
allocation slabs at least as large as our maximum buffer size though.
bq. You mean it is actually possible to implement the slicing without locks,
just with an atomic pointer? Quite possibly I guess.
Yes, and we do this already elsewhere. Here there is an issue with the
differing lifetimes of the buffers, though, and the time at which we recycle.
The ThreadLocal pools solve this problem by letting us reclaim together
bq. When the segmented file is closed or when the read command is completed?
When the read command has completed. We close all of our readers at the same
time, so we can release all of the slabs at once. Of course we could just have
a counter that is decremented on each close and relinquish them all together
that way, to keep our changes isolated.
> Remove FileCacheService, instead pooling the buffers
> ----------------------------------------------------
>
> Key: CASSANDRA-8897
> URL: https://issues.apache.org/jira/browse/CASSANDRA-8897
> Project: Cassandra
> Issue Type: Improvement
> Components: Core
> Reporter: Benedict
> Assignee: Stefania
> Fix For: 3.0
>
>
> After CASSANDRA-8893, a RAR will be a very lightweight object and will not
> need caching, so we can eliminate this cache entirely. Instead we should have
> a pool of buffers that are page-aligned.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
