[
https://issues.apache.org/jira/browse/IGNITE-2654?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15155481#comment-15155481
]
Ilya Lantukh edited comment on IGNITE-2654 at 2/20/16 8:38 AM:
---------------------------------------------------------------
I implemented similar optimizations for all mini futures that are created in
our benchmark. The more nodes are participating in each transaction, the higher
is relative difference. Here are results for 8 nodes in topology and 8 ops per
transaction.
Before:
||Benchmark|| Mode|| Cnt||
Score|| Error|| Units||
|JmhCacheBenchmark.put| thrpt| 60|
1268.916| ± 28.743| ops/s|
|JmhCacheBenchmark.put:·gc.alloc.rate| thrpt| 60|
144.218| ± 8.841| MB/sec|
|JmhCacheBenchmark.put:·gc.alloc.rate.norm| thrpt| 60|
119766.066| ± 5106.417| B/op|
|JmhCacheBenchmark.put:·gc.churn.PS_Eden_Space| thrpt| 60|
158.229| ± 144.091| MB/sec|
|JmhCacheBenchmark.put:·gc.churn.PS_Eden_Space.norm| thrpt| 60|
137731.820| ± 126813.796| B/op|
|JmhCacheBenchmark.put:·gc.churn.PS_Survivor_Space| thrpt| 60|
2.448| ± 5.902| MB/sec|
|JmhCacheBenchmark.put:·gc.churn.PS_Survivor_Space.norm| thrpt| 60|
2181.892| ± 5276.305| B/op|
|JmhCacheBenchmark.put:·gc.count| thrpt| 60|
12.000| | counts|
|JmhCacheBenchmark.put:·gc.time| thrpt| 60|
459.000| | ms|
After:
||Benchmark|| Mode|| Cnt||
Score|| Error|| Units||
|JmhCacheBenchmark.put| thrpt| 60|
1260.800| ± 25.131| ops/s|
|JmhCacheBenchmark.put:·gc.alloc.rate| thrpt| 60|
141.772| ± 8.169| MB/sec|
|JmhCacheBenchmark.put:·gc.alloc.rate.norm| thrpt| 60|
118371.266| ± 4963.497| B/op|
|JmhCacheBenchmark.put:·gc.churn.PS_Eden_Space| thrpt| 60|
147.477| ± 138.663| MB/sec|
|JmhCacheBenchmark.put:·gc.churn.PS_Eden_Space.norm| thrpt| 60|
136030.233| ± 124344.704| B/op|
|JmhCacheBenchmark.put:·gc.churn.PS_Survivor_Space| thrpt| 60|
2.587| ± 5.421| MB/sec|
|JmhCacheBenchmark.put:·gc.churn.PS_Survivor_Space.norm| thrpt| 60|
2258.795| ± 4756.145| B/op|
|JmhCacheBenchmark.put:·gc.count| thrpt| 60|
12.000| | counts|
|JmhCacheBenchmark.put:·gc.time| thrpt| 60|
460.000| | ms|
was (Author: ilantukh):
I implemented similar optimizations for all mini futures that are created in
our benchmark. The more nodes are participating in each transaction, the higher
is difference. Here are results for 8 nodes in topology and 8 ops per
transaction.
Before:
||Benchmark|| Mode|| Cnt||
Score|| Error|| Units||
|JmhCacheBenchmark.put| thrpt| 60|
1268.916| ± 28.743| ops/s|
|JmhCacheBenchmark.put:·gc.alloc.rate| thrpt| 60|
144.218| ± 8.841| MB/sec|
|JmhCacheBenchmark.put:·gc.alloc.rate.norm| thrpt| 60|
119766.066| ± 5106.417| B/op|
|JmhCacheBenchmark.put:·gc.churn.PS_Eden_Space| thrpt| 60|
158.229| ± 144.091| MB/sec|
|JmhCacheBenchmark.put:·gc.churn.PS_Eden_Space.norm| thrpt| 60|
137731.820| ± 126813.796| B/op|
|JmhCacheBenchmark.put:·gc.churn.PS_Survivor_Space| thrpt| 60|
2.448| ± 5.902| MB/sec|
|JmhCacheBenchmark.put:·gc.churn.PS_Survivor_Space.norm| thrpt| 60|
2181.892| ± 5276.305| B/op|
|JmhCacheBenchmark.put:·gc.count| thrpt| 60|
12.000| | counts|
|JmhCacheBenchmark.put:·gc.time| thrpt| 60|
459.000| | ms|
After:
||Benchmark|| Mode|| Cnt||
Score|| Error|| Units||
|JmhCacheBenchmark.put| thrpt| 60|
1260.800| ± 25.131| ops/s|
|JmhCacheBenchmark.put:·gc.alloc.rate| thrpt| 60|
141.772| ± 8.169| MB/sec|
|JmhCacheBenchmark.put:·gc.alloc.rate.norm| thrpt| 60|
118371.266| ± 4963.497| B/op|
|JmhCacheBenchmark.put:·gc.churn.PS_Eden_Space| thrpt| 60|
147.477| ± 138.663| MB/sec|
|JmhCacheBenchmark.put:·gc.churn.PS_Eden_Space.norm| thrpt| 60|
136030.233| ± 124344.704| B/op|
|JmhCacheBenchmark.put:·gc.churn.PS_Survivor_Space| thrpt| 60|
2.587| ± 5.421| MB/sec|
|JmhCacheBenchmark.put:·gc.churn.PS_Survivor_Space.norm| thrpt| 60|
2258.795| ± 4756.145| B/op|
|JmhCacheBenchmark.put:·gc.count| thrpt| 60|
12.000| | counts|
|JmhCacheBenchmark.put:·gc.time| thrpt| 60|
460.000| | ms|
> Protocol optimization for GridNearLockRequest/Response
> ------------------------------------------------------
>
> Key: IGNITE-2654
> URL: https://issues.apache.org/jira/browse/IGNITE-2654
> Project: Ignite
> Issue Type: Sub-task
> Components: general
> Reporter: Ilya Lantukh
> Assignee: Ilya Lantukh
> Fix For: 1.6
>
>
> Create new, more lightweight versions of GridNearLockRequest/Response:
> - Make miniId integer.
> - Store boolean flags in a single byte field.
> - Remove unused fields.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
