[
https://issues.apache.org/jira/browse/BOOKKEEPER-593?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13645285#comment-13645285
]
Sijie Guo commented on BOOKKEEPER-593:
--------------------------------------
from my understanding, this jira is a bit different from BOOKKEEPER-592, it
requires a new distribute schedule more than an ensemble selection.
for example, suppose A, B, C is the ensemble (A is the application machine),
write quorum size is 2. in round-robin way, it should 0-(A,B), 1-(B,C),
2-(C,A). in write-local-if-possible way, the write might be 0-(A,B), 1-(A, C),
2-(A, B).
so BOOKKEEPER-592 requires an pluggable ensemble selection policy. based on
BOOKKEEPER-592, an rack-aware ensemble policy could be provided, which returns
an ensemble whose first bookie is treated as the closest bookie with the
application. we need to provide a distributed schedule, which pick up the first
bookie (closest bookie) in write quorums. since distributed schedule is bound
how we wrote entries, distributed schedule should be recorded with ledger
metadata.
BTW, I think we could also use the locality for reads, which pick the closest
bookie as the starting point to read.
> use the application machine as one of bookies for ledger write if possible
> --------------------------------------------------------------------------
>
> Key: BOOKKEEPER-593
> URL: https://issues.apache.org/jira/browse/BOOKKEEPER-593
> Project: Bookkeeper
> Issue Type: Improvement
> Reporter: Ming Ma
> Assignee: Sijie Guo
>
> In the scenario where bk applications run on the same set of machines as
> bookies, it will be useful if bk can put one of the ledger replica on the
> same machine as the application. In that way one of the ledger writes will be
> local and could improve the performance.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
