GitHub - ergo-services/benchmarks: Benchmarks of the Ergo Framework 3.0 (and above) (original) (raw)

Benchmarks of the Ergo Framework 3.0 (and above)

The tests below are performed on the laptop Macbook Air M3 (2024)

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