You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@ignite.apache.org by "Aleksey Plekhanov (Jira)" <ji...@apache.org> on 2024/01/09 18:19:00 UTC
[jira] [Created] (IGNITE-21225) Redundant lambda object allocation in ClockPageReplacementFlags#setFlag
Aleksey Plekhanov created IGNITE-21225:
------------------------------------------
Summary: Redundant lambda object allocation in ClockPageReplacementFlags#setFlag
Key: IGNITE-21225
URL: https://issues.apache.org/jira/browse/IGNITE-21225
Project: Ignite
Issue Type: Improvement
Reporter: Aleksey Plekhanov
Assignee: Aleksey Plekhanov
Every time we call {{ClockPageReplacementFlags#setFlag/clearFlag}} methods the new lambda object is created, since lambda is accessing the variable in enclosing scope. \{{ClockPageReplacementFlags#setFlag}} method called every time when page is modified, so, it's a relatevily hot method and we should avoid new object allocation here.
Here is the test to show redundant allocations:
{code:java}
/** */
@Test
public void testAllocation() {
clockFlags = new ClockPageReplacementFlags(MAX_PAGES_CNT, region.address());
int cnt = 1_000_000;
ThreadMXBean bean = (ThreadMXBean)ManagementFactory.getThreadMXBean();
// Warmup.
clockFlags.setFlag(0);
long allocated0 = bean.getThreadAllocatedBytes(Thread.currentThread().getId());
for (int i = 0; i < cnt; i++)
clockFlags.setFlag(i % MAX_PAGES_CNT);
long allocated1 = bean.getThreadAllocatedBytes(Thread.currentThread().getId());
assertTrue("Too many bytes allocated: " + (allocated1 - allocated0), allocated1 - allocated0 < cnt);
} {code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)