Luo Chen has posted comments on this change.

Change subject: [ASTERIXDB-2276][CONF] Introduce Max Active Writable Datasets
......................................................................


Patch Set 1:

(1 comment)

https://asterix-gerrit.ics.uci.edu/#/c/2362/1//COMMIT_MSG
Commit Message:

PS1, Line 19: - Introduce max active writable datasets and use
            :   it to automatically calculate the number of pages
            :   of a user dataset memory component.
> as a user, would you rather configure a number related to datasets - someth
Hard to say. For me, I would rather configure storage.memorycomponent.numpages 
by myself to have more control over the system performance. It's meaningless to 
write many datasets at the same time without a performance guarantee.

This parameter is one key knob for ingestion performance. Setting max active 
number of datasets might seem to be "user-friendly", but it rather leaves the 
storage.memorycomponent.numpages as a black box, since we have to account for 
datasets, partitions, secondary indexes etc, to compute the actual num pages 
paramter. It's not quite straightforward what is the actual size of each memory 
component. Though the current configuration has the similar problem.

Another issue with max active writable datasets is that this parameter may 
cause some confusion. Say I have 10GB memory for global memory components, and 
I set this number to 100. However, if I only have 1 active dataset for write, 
what the system should behave? A "smart" system might take as much memory 
resources as possible to improve the ingestion performance to this only active 
writing dataset. However, this is not the behavior of our current system.


-- 
To view, visit https://asterix-gerrit.ics.uci.edu/2362
To unsubscribe, visit https://asterix-gerrit.ics.uci.edu/settings

Gerrit-MessageType: comment
Gerrit-Change-Id: I9909c26b1e12b431f913e201d2c3d83769be7269
Gerrit-PatchSet: 1
Gerrit-Project: asterixdb
Gerrit-Branch: master
Gerrit-Owner: Murtadha Hubail <mhub...@apache.org>
Gerrit-Reviewer: Anon. E. Moose #1000171
Gerrit-Reviewer: Jenkins <jenk...@fulliautomatix.ics.uci.edu>
Gerrit-Reviewer: Luo Chen <cl...@uci.edu>
Gerrit-Reviewer: Murtadha Hubail <mhub...@apache.org>
Gerrit-HasComments: Yes

Reply via email to