[
https://issues.apache.org/jira/browse/HDFS-14356?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16852707#comment-16852707
]
Feilong He commented on HDFS-14356:
-----------------------------------
I have some thoughts on pmem map address. As you know, NativePmemMappedBlock
keeps map address which can be used for cache read and unmap. This address can
also be obtained by calling pmem_map_file() of libpmem with cached file path
given. I am not sure how much performance drop can lead to by calling native
pmem_map_file() method instead. If performance drop is negligible, we can
consider to use this method, thus, NativePmemMappedBlock has no need to keep
map address.
If you have any suggestion, please post it on this page. Thanks!
> Implement HDFS cache on SCM with native PMDK libs
> -------------------------------------------------
>
> Key: HDFS-14356
> URL: https://issues.apache.org/jira/browse/HDFS-14356
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Components: caching, datanode
> Reporter: Feilong He
> Assignee: Feilong He
> Priority: Major
> Attachments: HDFS-14356.000.patch, HDFS-14356.001.patch,
> HDFS-14356.002.patch, HDFS-14356.003.patch, HDFS-14356.004.patch,
> HDFS-14356.005.patch, HDFS-14356.006.patch, HDFS-14356.007.patch
>
>
> In this implementation, native PMDK libs are used to map HDFS blocks to SCM.
> To use this implementation, user should build hadoop with PMDK libs by
> specifying a build option. This implementation is only supported on linux
> platform.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
