Benchmarks

Note

Benchmarks are hard.

Repeatedly calling the same function in a tight loop will lead to the instruction cache staying hot and branches being highly predictable. That’s not representative of real world access patterns. It’s also hard to write a nonbiased benchmark. I wrote msgspec, naturally whatever benchmark I publish it’s going to perform well in.

Even so, people like to see benchmarks. I’ve tried to be as nonbiased as I can be, and the results hopefully indicate a few tradeoffs you make when you choose different serialization formats. I encourage you to write your own benchmarks before making these decisions.

In all cases benchmarks were run on my local development machine (a ~2020 x86 Linux laptop) using CPython 3.11.

JSON Serialization & Validation

This benchmark covers the common case when working with msgspec or other validation libraries. It measures two things:

  • Decoding some JSON input, validating it against a schema, and converting it into user-friendly python objects.

  • Encoding these same python objects back into JSON.

The data we’re working with has the following schema (defined here using msgspec.Struct types):

import enum
import datetime
import msgspec

class Permissions(enum.Enum):
    READ = "READ"
    WRITE = "WRITE"
    READ_WRITE = "READ_WRITE"


class File(msgspec.Struct, kw_only=True, tag="file"):
    name: str
    created_by: str
    created_at: datetime.datetime
    updated_by: str | None = None
    updated_at: datetime.datetime | None = None
    nbytes: int
    permissions: Permissions


class Directory(msgspec.Struct, kw_only=True, tag="directory"):
    name: str
    created_by: str
    created_at: datetime.datetime
    updated_by: str | None = None
    updated_at: datetime.datetime | None = None
    contents: list[File | Directory]

The libraries we’re comparing are the following:

Each benchmark creates a message containing one or more File/Directory instances, then then serializes, deserializes, and validates it in a loop.

The full benchmark source can be found here.

In this benchmark msgspec is ~6x faster than mashumaro, ~10x faster than cattrs, and ~12x faster than pydantic V2, and ~85x faster than pydantic V1.

This plot shows the performance benefit of performing type validation during message decoding (as done by msgspec) rather than as a secondary step with a third-party library like cattrs or pydantic V1. Validating after decoding is slower for two reasons:

  • It requires traversing over the entire output structure a second time (which can be slow due to pointer chasing)

  • It may require converting some python objects to their desired output types (e.g. converting a decoded dict to a pydantic model), resulting in allocating many temporary python objects.

In contrast, libraries like msgspec that validate during decoding have none of these issues. Only a single pass over the decoded data is taken, and the specified output types are created correctly the first time, avoiding the need for additional unnecessary allocations.

This benefit also shows up in the memory usage for the same benchmark:

Here we compare the peak increase in memory usage (RSS) after loading the schemas and data. msgspec’s small library size, schema representation, and in-memory state means it uses a fraction of the memory of other tools.

JSON Serialization

msgspec includes its own high performance JSON library, which may be used by itself as a replacement for the standard library’s json.dumps/json.loads functions. Here we compare msgspec’s JSON implementation against several other popular Python JSON libraries.

The full benchmark source can be found here.

In this case msgspec structs (which measures msgspec with msgspec.Struct schemas pre-defined) is the fastest. When used without schemas, msgspec is on-par with orjson (the next fastest JSON library).

This shows that msgspec is able to decode JSON faster when a schema is provided. Due to a more efficient in memory representation, JSON decoding AND schema validation with msgspec than just JSON decoding alone.

MessagePack Serialization

Likewise, msgspec includes its own high performance MessagePack library, which may be used by itself without requiring usage of any of msgspec’s validation machinery. Here we compare msgspec’s MessagePack implementation against several other popular Python MessagePack libraries.

As with the JSON benchmark above, msgspec with a schema provided (msgspec structs) is faster than msgspec with no schema. In both cases though msgspec is measurably faster than other Python MessagePack libraries like msgpack or ormsgpack.

JSON Serialization - Large Data

Here we benchmark loading a large JSON file (~77 MiB) containing information on all the noarch packages in conda-forge. We compare the following libraries:

For each library, we measure both the peak increase in memory usage (RSS) and the time to JSON decode the file.

The full benchmark source can be found here.

Results (smaller is better):

memory (MiB)

vs.

time (ms)

vs.

msgspec structs

67.6

1.0x

176.8

1.0x

msgspec

218.3

3.2x

630.5

3.6x

json

295.0

4.4x

868.6

4.9x

ujson

349.1

5.2x

1087.0

6.1x

rapidjson

375.0

5.6x

1004.0

5.7x

orjson

406.3

6.0x

691.7

3.9x

simdjson

603.2

8.9x

1053.0

6.0x

  • msgspec decoding into Struct types uses the least amount of memory, and is also the fastest to decode. This makes sense; Struct types are cheaper to allocate and more memory efficient than dict types, and for large messages these differences can really add up.

  • msgspec decoding without a schema is the second best option for both memory usage and speed. When decoding without a schema, msgspec makes the assumption that the underlying message probably still has some structure; short dict keys are temporarily cached to be reused later on, rather than reallocated every time. This means that instead of allocating 10,000 copies of the string "name", only a single copy is allocated and reused. For large messages this can lead to significant memory savings. json and orjson also use similar optimizations, but not as effectively.

  • orjson and simdjson use 6-9x more memory than msgspec in this benchmark. In addition to the reasons above, both of these decoders require copying the original message into a temporary buffer. In this case, the extra copy adds an extra 77 MiB of overhead!

Structs

Here we benchmark common msgspec.Struct operations, comparing their performance against other similar libraries. The cases compared are:

For each library, the following operations are benchmarked:

  • Time to define a new class. Many libraries that abstract away class boilerplate add overhead when defining classes, slowing import times for libraries that make use of these classes.

  • Time to create an instance of that class.

  • Time to compare two instances for equality (==/!=).

  • Time to compare two instances for order (</>/<=/>=)

The full benchmark source can be found here.

Results (smaller is better):

import (μs)

create (μs)

equality (μs)

order (μs)

msgspec

12.51

0.09

0.02

0.03

standard classes

7.88

0.35

0.08

0.16

attrs

483.10

0.37

0.14

1.87

dataclasses

506.09

0.36

0.14

0.16

pydantic

673.47

1.54

0.60

N/A

  • Standard Python classes are the fastest to import (any library can only add overhead here). Still, msgspec isn’t that much slower, especially compared to other options.

  • Structs are optimized to be cheap to create, and that shows for the creation benchmark. They’re roughly 4x faster than standard classes/attrs/dataclasses, and 17x faster than pydantic.

  • For equality comparison, msgspec Structs are roughly 4x to 30x faster than the alternatives.

  • For order comparison, msgspec Structs are roughly 5x to 60x faster than the alternatives.

Garbage Collection

msgspec.Struct instances implement several optimizations for reducing garbage collection (GC) pressure and decreasing memory usage. Here we benchmark structs (with and without gc=False) against standard Python classes (with and without __slots__).

For each option we create a large dictionary containing many simple instances of the benchmarked type, then measure:

  • The amount of time it takes to do a full garbage collection (gc) pass

  • The total amount of memory used by this data structure

The full benchmark source can be found here.

Results (smaller is better):

GC time (ms)

Memory Used (MiB)

standard class

80.46

211.66

standard class with __slots__

80.06

120.11

msgspec struct

13.96

120.11

msgspec struct with gc=False

1.07

104.85

  • Standard Python classes are the most memory hungry (since all data is stored in an instance dict). They also result in the largest GC pause, as the GC has to traverse the entire outer dict, each class instance, and each instance dict. All that pointer chasing has a cost.

  • Standard classes with __slots__ are less memory hungry, but still results in an equivalent GC pauses.

  • msgspec.Struct instances have the same memory layout as a class with __slots__ (and thus have the same memory usage), but due to deferred GC tracking a full GC pass completes in a fraction of the time.

  • msgspec.Struct instances with gc=False have the lowest memory usage (lack of GC reduces memory by 16 bytes per instance). They also have the lowest GC pause (75x faster than standard classes!) since the entire composing dict can be skipped during GC traversal.

Library Size

Here we compare the on-disk size of msgspec and pydantic, its closest equivalent.

The full benchmark source can be found here.

Results (smaller is better)

version

size (MiB)

vs. msgspec

msgspec

0.18.4

0.46

1.00x

pydantic

2.5.2

6.71

14.66x

For applications where dependency size matters, msgspec is roughly 15x smaller on disk.