[
https://issues.apache.org/jira/browse/HDFS-14402?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16847482#comment-16847482
]
Rakesh R commented on HDFS-14402:
---------------------------------
bq. Upon making checksum configurable, I am thinking maybe to user, the cache
read performance is concerned mostly. The checksum is executed once when
caching data to DRAM/Pmem. It may be tolerable to user with checksum operation
for data verification. I think more discussions are required. I will open a
separate Jira common to DRAM and Pmem cache.
Yes, it makes sense to me to move the discussion to a separate jira. One use
case I can foresee that, if some user has concern about the sanity of data
while {{read-from-cache}} block then they will do checksum computation again at
read call. This is again depends on the NVMe device consistency level, I think.
In that case the checksum we did at the beginning can be skipped and not
worried about the sanity during {{write-to-cache}}.
> Use FileChannel.transferTo() method for transferring block to SCM cache
> -----------------------------------------------------------------------
>
> Key: HDFS-14402
> URL: https://issues.apache.org/jira/browse/HDFS-14402
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Components: caching, datanode
> Reporter: Feilong He
> Assignee: Feilong He
> Priority: Major
> Attachments: HDFS-14402.000.patch, HDFS-14402.001.patch,
> HDFS-14402.002.patch, With-Cache-Improvement-Patch.png,
> Without-Cache-Improvement-Patch.png
>
>
> We will consider to use transferTo API to improve SCM's cach performace.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
