[
https://issues.apache.org/jira/browse/KUDU-589?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Grant Henke resolved KUDU-589.
------------------------------
Fix Version/s: NA
Resolution: Won't Do
> Present our WAL benchmark findings
> ----------------------------------
>
> Key: KUDU-589
> URL: https://issues.apache.org/jira/browse/KUDU-589
> Project: Kudu
> Issue Type: Task
> Components: documentation
> Affects Versions: Backlog
> Reporter: Jean-Daniel Cryans
> Priority: Trivial
> Fix For: NA
>
>
> The [Disk scaling latency/throughput
> measurements|https://docs.google.com/a/cloudera.com/spreadsheets/d/1hnZQOxOo5Eti5oYjBnqpAuRGbw2awGcCvSdEEEePSg0/edit?usp=sharing]
> that we've been running since last week have taught us a lot about disks,
> filesystems, and fsync VS fdatasync.
> We should:
> # Present our findings to the rest of Eng. Even if it's not actionable right
> away, it can serve as a reference in the future.
> # Consider publishing our findings on the blog. It should attract systems
> engineers.
> Regarding the latter, I'm thinking that we would also have to:
> - Open source the wal benchmarking tool, scrub all the kudu-ness from it.
> - Run tests on some more hardware, so that we have enough meat (would be good
> to do enterprise-grade SSDs that's not Fusion-io, 10k rpms). Also, we should
> consider testing ext3.
> - Augment the benchmark so that it's more realistic, like testing multiple
> WALs, deletes, reading (basically simulate compactions), etc.
> I don't think we have to have the perfect blog post in one shot, we could do
> a second part if there's interest.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
