SSE vs Websocket Speed

[weoiss1998]

Platform

ESP32

IDE / Tooling

Arduino (IDE/CLI)

What happened?

Hi,

is the new update also affecting the SSE speed?
Which system should I use for a high repetition rate?
I couldn't find a forum or Discord, so I wanted to ask here.

Stack Trace

x

Minimal Reproductible Example (MRE)

x

I confirm that:

• I have read the documentation.
• I have searched for similar discussions.
• I have searched for similar issues.
• I have looked at the examples.
• I have upgraded to the lasted version of ESPAsyncWebServer (and AsyncTCP for ESP32).

[mathieucarbou]

Hi,

please read what is written before opening an issue!
here this is a forum question so I have moved the question.

The last updates do not impact the SSE code. What so you want to know and do exactly ? You question is not clear enough...

[weoiss1998]

Sorry. I was blind and overread it.

I would like to send updates to the client at irregular intervals. These updates may be close together in time. The changelog for version 3.10.0 states that WebSockets are now extremely fast. This raises the question for me: are WebSockets now faster than SSE?

[mathieucarbou]

Sorry. I was blind and overread it.

I would like to send updates to the client at irregular intervals. These updates may be close together in time. The changelog for version 3.10.0 states that WebSockets are now extremely fast. This raises the question for me: are WebSockets now faster than SSE?

Hi,

Yes WebSocket will always perform better than SSE and can even support binary. If speed is a concern, use WebSocket + binary frames. For example if you send Json data you can use MessagePack to serialize with ArduinoJson as MessagePack binary that you deserialize on client-side. That's what I did in the ESP-DASH library to improve its perf. WS is also bi-directional com.

SSE is only in the case for simple webapps who only need to receive events from server to client (sse is only from server to client). Com is slower and client to server com needs to be done with xhr requests then (slower and heavier also).

So it all depends on what you are building.

FYI, ESPHome web page on a MCU (if you know ESPHome) is using SSE. And they have plenty of events. SSE is not slow.

WS can just achieve higher transmission rates.

[weoiss1998]

Thanks for your reply. My plan is to send data from the server to the client, the client will not answer. The thing is, I want to refresh as fast as possible when new data is available.
As I want to use JSON-files, you would recommend WS?

[mathieucarbou]

From what you say, both would work. You did not specify clearly anything about how much data and how much frequent or how many clients...

[weoiss1998]

For the start it will be 1 client, maybe expanded in the future. I did a quick calculation and see that the payload is approximately 12KB. The interval between two updates can be in the low double-digit millisecond range.

[mathieucarbou]

12K, that's a lot and is not a short message. And you are impacted by TCP fragmentation here.

If it is Json, Json takes a lot of size when sent as text. If you serialize in binary with MessagePack, it will take half or a third of the size depending on your keys.

You seem to have pretty much the same problem I solved in the ESP-DASH library so you can look at how it is implemented.
We are building a big Json in memory, the problem is that this Json can be big, both for the heap and also to be sent, because when you serialize a Json, you have at one point in memory both object: the ArduinoJson object, plus the buffer. So this takes twice the memory actually. So to work around. while we build the json we also keep track of its size and when we reach a threshold, we send the buffer and continue building the rest of the json message. We send the json in several parts: first part (at most 4k if I remember), then second part, etc. And for each part, we serialize with message pack to make sure we can squeeze as much data as possible in one go.

With ESP-DASH I can reach more than 3-4 updates per second with graph refresh (several points per graph) so that's quite an average message size.

If your 12K is your compressed message pack size and you need to send 12K at 10ms delay, then I think this is not possible. But I never tried.... You would need to. But if this is your Json size, then that might be possible yes.

if you have one of your 12k json you can have an idea of their size if serialized as messagepack

• https://msgpack.solder.party
• https://jsontoolhub.com/msgpack-to-json

FYI, I know a user is sending 60 short messages per second without issue.