Benchmarks of the Ergo Framework 3.0 (and above)

The tests below are performed on the Apple M4 Max

Ping

Performs 4 scenarios:

• 1 process spawns 'pong'-process locally and sends 3M messages
• N processes spawn 'pong'-process locally and send 1M messages (N = number of CPU)
• 1 process spawns 'pong'-process on a remote node and sends 3M messages
• N processes spawn 'pong'-process on a remote node and send 1M messages (N = number of CPU)

Memory usage (per process)

Performs the following scenario:

• Takes node information that includes memory usage value.
• Starts 1M processes
• Takes node information 3 times with 1s intervals to make sure the GC has freed unused memory

Distributed Pub/Sub (1M subscribers)

Demonstrates event delivery performace by publishing 1 event from 1 producer to 1,000,000 subscribers distributed across 10 nodes.

Key optimization: Instead of 1,000,000 network messages (one per subscriber), only 10 network messages are sent (one per consumer node). Each node locally distributes to its 100K subscribers.

Results:

• Total subscribers: 1,000,000
• Producer nodes: 1
• Consumer nodes: 10
• Time to deliver all: 342ms
• Network messages sent: 10
• Delivery rate: 2.9M msg/sec

Run with go run .

Hardware: Apple M4 Max

Serialization benchmarks: EDF vs Protobuf vs Gob

These benchmarks compare EDF, EDF (+cache), Protobuf, and Gob serialization performance across common data types.

• EDF and Gob rely on runtime reflection, which dynamically inspects and serializes data structures at runtime
• Protobuf uses code generation, producing static type-safe marshalling and unmarshalling logic.

Data Type	EDF	EDF (+cache)	Protobuf	Gob	Winner	EDF Advantage
String Encode	29.96ns, 53B, 0a	23.45ns, 0B, 0a	44.37ns, 32B, 1a	76.42ns, 16B, 1a	EDF+Cache	47% faster than Protobuf, 69% faster than Gob
String Decode	76.63ns, 72B, 4a	76.62ns, 72B, 4a	67.45ns, 96B, 2a	429.5ns, 1000B, 19a	Protobuf	EDF 14% slower, but 6x faster than Gob
Map Encode	310.5ns, 325B, 10a	224.7ns, 204B, 5a	339.9ns, 112B, 5a	234.0ns, 32B, 2a	EDF+Cache	34% faster than Protobuf, competitive with Gob
Map Decode	557.4ns, 955B, 25a	468.8ns, 846B, 22a	353.3ns, 528B, 13a	6733ns, 8256B, 185a	Protobuf	EDF+Cache 33% slower, but 14x faster than Gob
Complex Struct Encode	277.8ns, 307B, 6a	269.8ns, 306B, 6a	474.3ns, 224B, 9a	357.1ns, 88B, 5a	EDF+Cache	43% faster than Protobuf, 24% faster than Gob
Complex Struct Decode	740.1ns, 1364B, 41a	700.8ns, 1368B, 41a	597.2ns, 824B, 25a	9335ns, 10872B, 255a	Protobuf	EDF+Cache 17% slower, but 13x faster than Gob
Nested Struct Encode	739.9ns, 732B, 17a	796.2ns, 846B, 17a	1557ns, 640B, 27a	901.4ns, 256B, 15a	EDF	52% faster than Protobuf, 18% faster than Gob
Nested Struct Decode	2137ns, 4594B, 107a	2291ns, 5438B, 120a	1684ns, 2544B, 71a	12729ns, 14712B, 342a	Protobuf	EDF 27% slower, but 6x faster than Gob

Format: time ns/op, memory B/op, allocations/op

Run with go test -bench=. -benchmem

Hardware: Apple M4 Max