hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Elek, Marton (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDDS-225) Provide docker-compose files to check the scalability of KSM
Date Thu, 05 Jul 2018 09:08:00 GMT

    [ https://issues.apache.org/jira/browse/HDDS-225?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16533422#comment-16533422
] 

Elek, Marton commented on HDDS-225:
-----------------------------------

I uploaded a WIP utility to generate ksm volumes/buckets/keys. I check an existing key (http://ksm/template/template/template)
and use the containerid/localid of that key to generate more keys.

I tested it with locally and in the cloud. Locally I can generate 18-20000 key/sec (tempfs!!!)
in the coud I can generate 14-15000 key/sec (ramfs!!!). I would like to generate a huge number
of the keys and upload the generated data to somewhere.

Space requirement of 727 700 000 keys is ~12Gb.

> Provide docker-compose files to check the scalability of KSM
> ------------------------------------------------------------
>
>                 Key: HDDS-225
>                 URL: https://issues.apache.org/jira/browse/HDDS-225
>             Project: Hadoop Distributed Data Store
>          Issue Type: Improvement
>            Reporter: Elek, Marton
>            Priority: Major
>         Attachments: HDDS-225.001.patch
>
>
> I open this jira to start a discussion. The main question: how can we prove the scalability
of KSM with minimal effort?
> 1. The goal is to prove that KSM could handle 1-10 billion of keys without any problem.
> 2. 10 000 000 000 * 10 kbyte object = 10 Terrabyte space. But we need to test only the
KSM part.
> 3. With a low level data generator we can generate the volumes/buckets/keys directly
to the ksm.db (rocksdb). We can fake the block allocation and use exactly the same containerid/localid
for all the keys. With this method we can test the read/list methods without any problems
(all of the keys could be downloaded.
> 4. With this storage optimization we can  test 10 billion keys locally with a specific
docker-compose setup where the local db-s are mounted from the local directory
> 5. The data could be generated (takes some time) or could be uploaded after a generation
>   



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-help@hadoop.apache.org


Mime
View raw message