[
https://issues.apache.org/jira/browse/CASSANDRA-19679?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17852490#comment-17852490
]
Sam Lightfoot edited comment on CASSANDRA-19679 at 6/5/24 3:07 PM:
-------------------------------------------------------------------
Using for-i slightly improves allocations, reducing from 1.69% to 1.37%.
The load is generated via easy-cass-stress
$ bin/easy-cass-stress run KeyValue -n 10m -r 0.5 --drop --rate 10000
--compaction twcs --max-connections 16
which performs single partition inserts & point queries
{code:java}
CREATE TABLE IF NOT EXISTS keyvalue (
key text PRIMARY KEY,
value text
) WITH compaction = {
'class' : 'TimeWindowCompactionStrategy'{code}
was (Author: JIRAUSER302824):
Using for-i slightly improves allocations, reducing from 1.69% to 1.37%
!image-2024-06-05-16-05-59-870.png|width=1327,height=969!
The load is generated via easy-cass-stress
$ bin/easy-cass-stress run KeyValue -n 10m -r 0.5 --drop --rate 10000
--compaction twcs --max-connections 16
which performs single partition inserts & point queries
{code:java}
CREATE TABLE IF NOT EXISTS keyvalue (
key text PRIMARY KEY,
value text
) WITH compaction = {
'class' : 'TimeWindowCompactionStrategy'{code}
> Stream processing for SimpleRestriction::bindAndGetClusteringElements
> ---------------------------------------------------------------------
>
> Key: CASSANDRA-19679
> URL: https://issues.apache.org/jira/browse/CASSANDRA-19679
> Project: Cassandra
> Issue Type: Improvement
> Components: Legacy/Core
> Reporter: Sam Lightfoot
> Assignee: Sam Lightfoot
> Priority: Normal
> Fix For: 5.x
>
> Attachments: after.png, alloc_after.html, alloc_before.html,
> image-2024-06-04-21-55-34-457.png, image-2024-06-05-16-05-59-870.png
>
> Time Spent: 3h 10m
> Remaining Estimate: 0h
>
> Part 2 (of 2) of low-hanging fruit Stream performance improvements.
> The second main Stream contributor to allocations and CPU was
> SimpleRestriction::bindAndGetClusteringElements, which contributes to 5% of
> all allocations for a 50/50 workload. The image attached shows allocation
> profiling on _trunk_ (see purple highlighted sections for Stream-related
> allocs).
> The 'after' profile for a 50/50 workload shows a reduction from 4.58%
> allocations down to 1.69%. For a 90/10 (w/r) workload we see 4.28% decrease
> to 1.10%.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
