Todd Lipcon has posted comments on this change. ( 
http://gerrit.cloudera.org:8080/5905 )

Change subject: KUDU-1865 (part 1): reduce some cross-thread allocations
......................................................................


Patch Set 7:

Without positive (statistically significant) perf results I'm not sure it's 
worth doing this patch (at least the std::function struct stuff which is 
somewhat "weird" looking).

I seem to recall that I did this work originally as part of the integration of 
KRPC into Impala, and Impala was suffering majorly from the "AllocLarge" 
bottleneck on older versions of tcmalloc (fixed in the versions we use today). 
Impala did a lot of >1M allocations which went throuhg the AllocLarge slow path 
while holding the page heap lock, so any time that the RPC threads had to go 
through the central free list, they were likely to get blocked behind some 
other thread holding that lock. This made the cross-thread allocations really 
bad.

Now with more modern tcmalloc, it's still preferable to avoid the central free 
lists, but I don't think it's worth being too crazy without showing a real 
benefit.


--
To view, visit http://gerrit.cloudera.org:8080/5905
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-Project: kudu
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: I7d4d5f14fb302196b1797c712b21cfce81f157c1
Gerrit-Change-Number: 5905
Gerrit-PatchSet: 7
Gerrit-Owner: Todd Lipcon <[email protected]>
Gerrit-Reviewer: Alexey Serbin <[email protected]>
Gerrit-Reviewer: Andrew Wong <[email protected]>
Gerrit-Reviewer: Bankim Bhavsar <[email protected]>
Gerrit-Reviewer: David Ribeiro Alves <[email protected]>
Gerrit-Reviewer: Grant Henke <[email protected]>
Gerrit-Reviewer: Kudu Jenkins (120)
Gerrit-Reviewer: Michael Ho <[email protected]>
Gerrit-Reviewer: Sailesh Mukil <[email protected]>
Gerrit-Reviewer: Tidy Bot (241)
Gerrit-Reviewer: Todd Lipcon <[email protected]>
Gerrit-Comment-Date: Wed, 27 May 2020 03:58:23 +0000
Gerrit-HasComments: No

Reply via email to