Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Save complex OS-level I/O to know the CPU topology #216

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Save complex OS-level I/O to know the CPU topology #216

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7661c28
Save complex OS-level I/O to know the CPU topology
franz1981 Jan 8, 2025
1d1403f
Address David comments
franz1981 Jan 8, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
39 changes: 23 additions & 16 deletions src/main/java/org/jboss/threads/EnhancedQueueExecutor.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,7 +47,6 @@
import javax.management.ObjectInstance;
import javax.management.ObjectName;

import io.smallrye.common.cpu.CacheInfo;
import org.jboss.threads.management.ManageableThreadPoolExecutorService;
import org.jboss.threads.management.StandardThreadPoolMXBean;

Expand Down Expand Up @@ -337,23 +336,31 @@ private static final class RuntimeFields {
private static final long queueSizeOffset;

static {
int cacheLine = CacheInfo.getSmallestDataCacheLineSize();
if (cacheLine == 0) {
// guess
cacheLine = 64;
}
// cpu spatial prefetcher can drag 2 cache-lines at once into L2
int pad = cacheLine > 128 ? cacheLine : 128;
int longScale = unsafe.arrayIndexScale(long[].class);
int taskNodeScale = unsafe.arrayIndexScale(TaskNode[].class);
// these fields are in units of array scale
unsharedTaskNodesSize = pad / taskNodeScale * (numUnsharedObjects + 1);
unsharedLongsSize = pad / longScale * (numUnsharedLongs + 1);
// these fields are in bytes
headOffset = unsafe.arrayBaseOffset(TaskNode[].class) + pad;
tailOffset = unsafe.arrayBaseOffset(TaskNode[].class) + pad * 2;
threadStatusOffset = unsafe.arrayBaseOffset(long[].class) + pad;
queueSizeOffset = unsafe.arrayBaseOffset(long[].class) + pad * 2;
if (ProcessorInfo.availableProcessors() > 1) {
// this is fine for pretty much 32 and 64 bit x86 and ARM processors; see
// https://github.com/ziglang/zig/blob/0.13.0/lib/std/atomic.zig#L424-L434
// cpu spatial prefetcher can drag 2 cache-lines at once into L2
int pad = 128;
// we both pad before and after the array to avoid false sharing with surrounding heap objects
unsharedTaskNodesSize = pad / taskNodeScale * (numUnsharedObjects + 1);
unsharedLongsSize = pad / longScale * (numUnsharedLongs + 1);
// these fields are in bytes
headOffset = unsafe.arrayBaseOffset(TaskNode[].class) + pad;
tailOffset = unsafe.arrayBaseOffset(TaskNode[].class) + pad * 2;
threadStatusOffset = unsafe.arrayBaseOffset(long[].class) + pad;
queueSizeOffset = unsafe.arrayBaseOffset(long[].class) + pad * 2;
} else {
unsharedTaskNodesSize = numUnsharedObjects;
Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@dmlloyd we have no coverage for this unless we run our test suite with -XX:ActiveProcessorCount=1
so please take a look if I did something wrong

unsharedLongsSize = numUnsharedLongs;
// position 0
headOffset = unsafe.arrayBaseOffset(TaskNode[].class);
// position 1 in the object array
tailOffset = unsafe.arrayBaseOffset(TaskNode[].class) + taskNodeScale;
threadStatusOffset = unsafe.arrayBaseOffset(long[].class);
queueSizeOffset = unsafe.arrayBaseOffset(long[].class) + longScale;
}
}
}

Expand Down
Loading