Benchmark Java Serialisation / Deserialization Speeds Against ActiveJ Serialisation #44

Activity

1. chronoxor commented on Feb 16, 2021

   

   chronoxor

   on Feb 16, 2021

   Last edited by chronoxor

   Owner

   More actions

   We're not competing with native language serializers, because we're cross platform, cross language and cross wire.

   The fastest serialize will be memcpy(sizeof(MyStruct)) in C++ :-)
2. sirinath commented on Feb 16, 2021

   

   sirinath

   on Feb 16, 2021

   Author

   More actions

   Colfer serialisation (https://github.com/pascaldekloe/colfer) is the fastest cross platform serialisation which you can use for benchmarking. But it would be worth while to see how you hold up with language specific implementations also.
3. chronoxor commented on Feb 16, 2021

   

   chronoxor

   on Feb 16, 2021

   Owner

   More actions

   It was easy to benchmark the same model as Colfer serialization does for testing the serialization performance.

   bench.fbe:

   package bench
   version 1.0
   
   struct Benchmark
   {
       int64 key;
       string host;
       int32 port;
       int64 sise;
       uint64 hash;
       double ratio;
       bool route;
   }
   

   serialize_bench.cpp:

   #include "benchmark/cppbenchmark.h"
   #include "../proto/bench_models.h"
   
   class SerializationFixture
   {
   protected:
       FBE::bench::BenchmarkModel writer;
       std::vector<bench::Benchmark> benchmarks;
       size_t index{0};
       size_t size{4};
   
       SerializationFixture()
       {
           benchmarks.emplace_back(bench::Benchmark(1234567890, "db003lz12", 389, 452, 0x488b5c2428488918, 0.99, true));
           benchmarks.emplace_back(bench::Benchmark(1234567891, "localhost", 22, 4096, 0x243048899c24c824, 0.20, false));
           benchmarks.emplace_back(bench::Benchmark(1234567892, "kdc.local", 88, 1984, 0x000048891c24485c, 0.06, false));
           benchmarks.emplace_back(bench::Benchmark(1234567893, "vhost8.dmz.example.com", 27017, 59741, 0x5c2408488b9c2489, 0.0, true));
       }
   };
   
   BENCHMARK_FIXTURE(SerializationFixture, "Serialize")
   {
       // Reset writer stream
       writer.reset();
   
       // Serialize
       writer.serialize(benchmarks[index++ % size]);
   }
   
   BENCHMARK_MAIN()
   

   deserialize_bench.cpp:

   #include "benchmark/cppbenchmark.h"
   #include "../proto/bench_models.h"
   
   class DeserializationFixture
   {
   protected:
       FBE::bench::BenchmarkModel writer;
       FBE::bench::BenchmarkModel reader;
       bench::Benchmark deserialized;
       size_t index{0};
       size_t size{4};
   
       DeserializationFixture()
       {
           std::vector<bench::Benchmark> benchmarks;
           benchmarks.emplace_back(bench::Benchmark(1234567890, "db003lz12", 389, 452, 0x488b5c2428488918, 0.99, true));
           benchmarks.emplace_back(bench::Benchmark(1234567891, "localhost", 22, 4096, 0x243048899c24c824, 0.20, false));
           benchmarks.emplace_back(bench::Benchmark(1234567892, "kdc.local", 88, 1984, 0x000048891c24485c, 0.06, false));
           benchmarks.emplace_back(bench::Benchmark(1234567893, "vhost8.dmz.example.com", 27017, 59741, 0x5c2408488b9c2489, 0.0, true));
   
           for (const auto& benchmark : benchmarks)
               writer.serialize(benchmark);
           assert(writer.verify() && "Model is broken!");
           reader.attach(writer.buffer());
           assert(reader.verify() && "Model is broken!");
       }
   };
   
   BENCHMARK_FIXTURE(DeserializationFixture, "Deserialize")
   {
       // Deserialize
       reader.deserialize(deserialized);
   
       ++index;
   
       // Reset reader stream
       if (index == size)
           reader.attach(writer.buffer());
   }
   
   BENCHMARK_MAIN()
   
4. chronoxor commented on Feb 16, 2021

   

   chronoxor

   on Feb 16, 2021

   Owner

   More actions

   Serialization benchmark results:

   Phase: Serialize
   Average time: 31 ns/op
   Minimal time: 37 ns/op
   Maximal time: 39 ns/op
   Total time: 2.667 s
   Total operations: 86009850
   Operations throughput: 32239885 ops/s
   

   Deserialization benchmark results:

   Phase: Deserialize
   Average time: 18 ns/op
   Minimal time: 21 ns/op
   Maximal time: 22 ns/op
   Total time: 1.871 s
   Total operations: 101784438
   Operations throughput: 54389865 ops/s
   

   👍React with 👍1Reacted by Suminda Sirinath Salpitikorala Dharmasena