Most WebSocket servers exchange JSON messages because they’re convenient to parse and serialize in a browser. These messages contain text data and tend to be repetitive.

This makes the stream of messages highly compressible. Enabling compression can reduce network traffic by more than 80%.

There’s a standard for compressing messages. RFC 7692 defines WebSocket Per-Message Deflate, a compression extension based on the Deflate algorithm.

Configuring compression

connect() and serve() enable compression by default because the reduction in network bandwidth is usually worth the additional memory and CPU cost.

If you want to disable compression, set compression=None:

import websockets

websockets.connect(..., compression=None)

websockets.serve(..., compression=None)

If you want to customize compression settings, you can enable the Per-Message Deflate extension explicitly with ClientPerMessageDeflateFactory or ServerPerMessageDeflateFactory:

import websockets
from websockets.extensions import permessage_deflate

            compress_settings={"memLevel": 4},

            compress_settings={"memLevel": 4},

The Window Bits and Memory Level values in these examples reduce memory usage at the expense of compression rate.

Compression settings

When a client and a server enable the Per-Message Deflate extension, they negotiate two parameters to guarantee compatibility between compression and decompression. These parameters affect the trade-off between compression rate and memory usage for both sides.

  • Context Takeover means that the compression context is retained between messages. In other words, compression is applied to the stream of messages rather than to each message individually.

    Context takeover should remain enabled to get good performance on applications that send a stream of messages with similar structure, that is, most applications.

    This requires retaining the compression context and state between messages, which increases the memory footprint of a connection.

  • Window Bits controls the size of the compression context. It must be an integer between 9 (lowest memory usage) and 15 (best compression). Setting it to 8 is possible but rejected by some versions of zlib.

    On the server side, websockets defaults to 12. On the client side, it lets the server pick a suitable value, which is the same as defaulting to 15.

zlib offers additional parameters for tuning compression. They control the trade-off between compression rate, memory usage, and CPU usage only for compressing. They’re transparent for decompressing. Unless mentioned otherwise, websockets inherits defaults of compressobj().

  • Memory Level controls the size of the compression state. It must be an integer between 1 (lowest memory usage) and 9 (best compression).

    websockets defaults to 5. This is lower than zlib’s default of 8. Not only does a lower memory level reduce memory usage, but it can also increase speed thanks to memory locality.

  • Compression Level controls the effort to optimize compression. It must be an integer between 1 (lowest CPU usage) and 9 (best compression).

  • Strategy selects the compression strategy. The best choice depends on the type of data being compressed.

Tuning compression

For servers

By default, websockets enables compression with conservative settings that optimize memory usage at the cost of a slightly worse compression rate: Window Bits = 12 and Memory Level = 5. This strikes a good balance for small messages that are typical of WebSocket servers.

Here’s how various compression settings affect memory usage of a single connection on a 64-bit system, as well a benchmark of compressed size and compression time for a corpus of small JSON documents.

Window Bits

Memory Level

Memory usage

Size vs. default

Time vs. default



322 KiB





178 KiB





106 KiB





70 KiB





52 KiB





43 KiB





39 KiB



19 KiB


Window Bits and Memory Level don’t have to move in lockstep. However, other combinations don’t yield significantly better results than those shown above.

Compressed size and compression time depend heavily on the kind of messages exchanged by the application so this example may not apply to your use case.

You can adapt compression/ by creating a list of typical messages and passing it to the _run function.

Window Bits = 11 and Memory Level = 4 looks like the sweet spot in this table.

websockets defaults to Window Bits = 12 and Memory Level = 5 to stay away from Window Bits = 10 or Memory Level = 3 where performance craters, raising doubts on what could happen at Window Bits = 11 and Memory Level = 4 on a different corpus.

Defaults must be safe for all applications, hence a more conservative choice.

The benchmark focuses on compression because it’s more expensive than decompression. Indeed, leaving aside small allocations, theoretical memory usage is:

  • (1 << (windowBits + 2)) + (1 << (memLevel + 9)) for compression;

  • 1 << windowBits for decompression.

CPU usage is also higher for compression than decompression.

For clients

By default, websockets enables compression with Memory Level = 5 but leaves the Window Bits setting up to the server.

There’s two good reasons and one bad reason for not optimizing the client side like the server side:

  1. If the maintainers of a server configured some optimized settings, we don’t want to override them with more restrictive settings.

  2. Optimizing memory usage doesn’t matter very much for clients because it’s uncommon to open thousands of client connections in a program.

  3. On a more pragmatic note, some servers misbehave badly when a client configures compression settings. AWS API Gateway is the worst offender.

    Unfortunately, even though websockets is right and AWS is wrong, many users jump to the conclusion that websockets doesn’t work.

    Until the ecosystem levels up, interoperability with buggy servers seems more valuable than optimizing memory usage.

Further reading

This blog post by Ilya Grigorik provides more details about how compression settings affect memory usage and how to optimize them.

This experiment by Peter Thorson recommends Window Bits = 11 and Memory Level = 4 for optimizing memory usage.