[
https://issues.apache.org/jira/browse/IGNITE-11857?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16989854#comment-16989854
]
Aleksey Plekhanov commented on IGNITE-11857:
--------------------------------------------
[~ascherbakov]
We can only change the type from {{long}} to {{int}} for delta values. There is
no boxing for delta values in the code.
Resident size for both structures is also relatively low.
{code:java}
System.out.println("Size: " +
GraphLayout.parseInstance(benchmark.partCntr).totalSize())
{code}
Old implementation: 1752 bytes
TreeMap<Long, Item> (long delta): 2248 bytes
TreeMap<Long, Item> (int delta): 2072 bytes
TreeMap<Long, Integer>: 1816 bytes
Even if we have 100 cache groups with 1000 partitions on each node it's about
30-50 extra megabytes of heap.
This test is synthetic, in real production cases I think there will be much
lower values.
TreeMap with int delta has a little drop compared to long deltas and map of
<Long, Integer> has a little drop compared to int deltas.
Overall comparison table:
||Implementation||Throughput||GC pressure||Footprint||
|Old implementation|4,338 ± 0,208 ops/us|25 182 056|1752|
|TreeMap<Long, Item> (long delta)|6,089 ± 0,459 ops/us|11 828 272|2248|
|TreeMap<Long, Item> (long delta)|5,905 ± 0,338 ops/us|11 436 760|2072|
|TreeMap<Long, Integer>|4,857 ± 0,318 ops/us|10 733 888|1816|
I think we should get "TreeMap<Long, Item> (long delta)" version. What do you
think?
> Investigate performance drop after IGNITE-10078
> -----------------------------------------------
>
> Key: IGNITE-11857
> URL: https://issues.apache.org/jira/browse/IGNITE-11857
> Project: Ignite
> Issue Type: Improvement
> Reporter: Alexei Scherbakov
> Assignee: Aleksey Plekhanov
> Priority: Major
> Attachments: ignite-config.xml,
> run.properties.tx-optimistic-put-b-backup
>
> Time Spent: 20m
> Remaining Estimate: 0h
>
> After IGNITE-10078 yardstick tests show performance drop up to 8% in some
> scenarios:
> * tx-optim-repRead-put-get
> * tx-optimistic-put
> * tx-putAll
> Partially this is due new update counter implementation, but not only.
> Investigation is required.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
