NativeAOT EventPipe drops events in multi-threaded scenarios.

[LakshanF]

NativeAOT EventPipe seems to drop significant number of events when sending events over multiple threads. Leveraging the EventPipeStress program shows results for different configuration values as shown in the table below.

• The efficiency column (2nd column) shows the ratio of events received and dropped (shown in the 3rd and 4th columns in millions of events)
• NativeAOT transmits more events than CoreCLR (roughly 3 times) and require a larger buffer size in the default mode to hit similar efficiency as CoreCLR
• Transmitting events in multiple threads shows significant event drops (50%) when the duration of the events gets longer (60 seconds)
• Stress program needs to be built as a NativeAOT application and EventSourceSupport needs to be set to true.

Runtime	Efficiency	EventsReceived (M)	EventsDropped (M)	BufferSize	Threads	EventRate (M)	EventCount (M)	BurstPattern	EventSize	Duration (s)
NativeAOT	52	176	163	32768	2	-1	-1	None	100	60
NativeAOT	50	5	5	4096	10	-1	10	None	100	2
NativeAOT	37	198	338	4096	10	1	-1	None	100	60
NativeAOT	51	51	48	4096	10	10	10	None	100	12
NativeAOT	68	143	66	256	1	-1	-1	None	100	60
CoreCLR	99	58	0.7	256	1	-1	-1	None	100	60
NativeAOT	95	182	9	4096	1	-1	-1	None	100	60
CoreCLR	100	58	0	4096	1	-1	-1	None	100	60
NativeAOT	91	191	18	32768	1	-1	-1	None	100	60
CoreCLR	100	103	0	32768	2	-1	-1	None	100	60
NativeAOT	100	10	0	4096	10	-1	1	None	100	2
NativeAOT	100	1	0	4096	10	10	0.1	None	100	0.2
NativeAOT	100	10	0	4096	10	10	1	None	100	2
NativeAOT	100	10	0	4096	10	10	1	BOLUS	100	2
NativeAOT	100	10	0	4096	10	10	1	DRIP	100	2
NativeAOT	100	10	0	4096	10	10	1	HEAVY_DRIP	100	2

Tagging subscribers to this area: @agocke, @MichalStrehovsky, @jkotas
See info in area-owners.md if you want to be subscribed.

Issue Details

---

NativeAOT EventPipe seems to drop significant number of events when sending events over multiple threads. Leveraging the EventPipeStress program shows results for different configuration values as shown in the table below.

• The efficiency column (2nd column) shows the ratio of events received and dropped (shown in the 3rd and 4th columns in millions of events)
• NativeAOT transmits more events than CoreCLR (roughly 3 times) and require a larger buffer size in the default mode to hit similar efficiency as CoreCLR
• Transmitting events in multiple threads shows significant event drops (50%) when the duration of the events gets longer (60 seconds)
• Stress program needs to be built as a NativeAOT application and EventSourceSupport needs to be set to true.

Runtime	Efficiency	EventsReceived (M)	EventsDropped (M)	BufferSize	Threads	EventRate (M)	EventCount (M)	BurstPattern	EventSize	Duration (s)
NativeAOT	52	176	163	32768	2	-1	-1	None	100	60
NativeAOT	50	5	5	4096	10	-1	10	None	100	2
NativeAOT	37	198	338	4096	10	1	-1	None	100	60
NativeAOT	51	51	48	4096	10	10	10	None	100	12
NativeAOT	68	143	66	256	1	-1	-1	None	100	60
CoreCLR	99	58	0.7	256	1	-1	-1	None	100	60
NativeAOT	95	182	9	4096	1	-1	-1	None	100	60
CoreCLR	100	58	0	4096	1	-1	-1	None	100	60
NativeAOT	91	191	18	32768	1	-1	-1	None	100	60
CoreCLR	100	103	0	32768	2	-1	-1	None	100	60
NativeAOT	100	10	0	4096	10	-1	1	None	100	2
NativeAOT	100	1	0	4096	10	10	0.1	None	100	0.2
NativeAOT	100	10	0	4096	10	10	1	None	100	2
NativeAOT	100	10	0	4096	10	10	1	BOLUS	100	2
NativeAOT	100	10	0	4096	10	10	1	DRIP	100	2
NativeAOT	100	10	0	4096	10	10	1	HEAVY_DRIP	100	2

Author:	LakshanF
Assignees:	-
Labels:	area-NativeAOT-coreclr
Milestone:	8.0.0

[LakshanF]

Given this is hitting with large number of events, and the apparent transmission spike seen with NativeAOT, I think this can be punted to .NET9