[ 
https://issues.apache.org/jira/browse/HDFS-7661?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15166441#comment-15166441
 ] 

Kai Zheng commented on HDFS-7661:
---------------------------------

I did a quick reading of the v2 design doc. Some comments and questions:
* Overall, I'm not sure why introducing a new meta file {{.bgLen}} for striped 
parity block is better, than augmenting the existing block meta file. Using a 
new meta file, it will means it has to stay along with the block where the 
block moves/replicates/reconstructs. Also, why just keep it for parity blocks? 
Maybe not bad for all the BG blocks.
* We may need well documenting about {{offsetInBlock, packetLen, 
blockGroupLen}} and why we need them. The names may be refined. Otherwise 
someone may wonder why such intermediate variables need to be persisted as part 
of meta data.
* bq. Consider the default EC policy whose cell size 65536B (64KB), and the 
DFSPacket data size is 64512B(63KB)
The assumption isn't good, because I don't think it's a good idea to have 
cell-size and packet-data-size like this, not multiplied. It's hard to align 
the buffer address for erasure encoding and checksum computing ({{both are 
performance critical}}) without buffer data copying. We should ensure either 
cell-size or packet-size can fall into the other, or for simple, they're equal.

I may have more comments in the following days, thanks for addressing or 
clarifying them.

> Erasure coding: support hflush and hsync
> ----------------------------------------
>
>                 Key: HDFS-7661
>                 URL: https://issues.apache.org/jira/browse/HDFS-7661
>             Project: Hadoop HDFS
>          Issue Type: Sub-task
>            Reporter: Tsz Wo Nicholas Sze
>            Assignee: GAO Rui
>         Attachments: EC-file-flush-and-sync-steps-plan-2015-12-01.png, 
> HDFS-7661-unitTest-wip-trunk.patch, HDFS-7661-wip.01.patch, 
> HDFS-EC-file-flush-sync-design-version1.1.pdf, 
> HDFS-EC-file-flush-sync-design-version2.0.pdf
>
>
> We also need to support hflush/hsync and visible length. 



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to