[
https://issues.apache.org/jira/browse/HBASE-13291?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14390088#comment-14390088
]
stack commented on HBASE-13291:
-------------------------------
bq. I think when we discussed on this API during tags it was decided that it
would be redundant as getTagsLength() would be enough.
Yeah, we probably said that.
I have rig w/ CPU maxed out and am trying to knock off the top consumers so we
can get more out when a hot box. In this pure scan scenario, all from
blockcache, the repeated parse of Cell lengths -- key, value, row, qualifier --
when we want to check pieces of it (row, family, etc.) adds up.
We could cache some of these calculations (as you do in your BB patch) but
while back we thought it not worth the bulk-up in the heap. I could retry it...
> Lift the scan ceiling
> ---------------------
>
> Key: HBASE-13291
> URL: https://issues.apache.org/jira/browse/HBASE-13291
> Project: HBase
> Issue Type: Improvement
> Components: Scanners
> Affects Versions: 1.0.0
> Reporter: stack
> Assignee: stack
> Attachments: 13291.hacks.txt, 13291.inlining.txt, Screen Shot
> 2015-03-26 at 12.12.13 PM.png, Screen Shot 2015-03-26 at 3.39.33 PM.png,
> hack_to_bypass_bb.txt, nonBBposAndInineMvccVint.txt, q (1).png, traces.7.svg,
> traces.filterall.svg, traces.nofilter.svg, traces.small2.svg,
> traces.smaller.svg
>
>
> Scanning medium sized rows with multiple concurrent scanners exhibits
> interesting 'ceiling' properties. A server runs at about 6.7k ops a second
> using 450% of possible 1600% of CPUs when 4 clients each with 10 threads
> doing scan 1000 rows. If I add '--filterAll' argument (do not return
> results), then we run at 1450% of possible 1600% possible but we do 8k ops a
> second.
> Let me attach flame graphs for two cases. Unfortunately, there is some
> frustrating dark art going on. Let me try figure it... Filing issue in
> meantime to keep score in.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
