mimalloc is deprecated

[StefanOltmann]

Can't build with that compiler arg and Kotlin 2.2

It throws an exception

[trueangle]

Hey @StefanOltmann

Thanks for the info. I will check that. Would be nice to see full exception message, could you share it?

I guess, for now, the only option is to use Kotlin (2.1.10) or try using different allocator

[StefanOltmann]

Hi @trueangle,

The exact message is this:

> Task :linkDebugExecutableLinuxX64 FAILED
5 actionable tasks: 5 executed
e: Usage of mimalloc in Kotlin/Native compiler is deprecated. Please remove -Xallocator=mimalloc compiler flag.

FAILURE: Build failed with an exception.

I'm interested in ways to further improve the performance of the Lambda function. For example, is Amazon Linux 2023 faster than Amazon Linux 2?

If you're wondering where all the recent stars for your great library came from, I might have an explanation. ;)

Thank you for your great work!

Kind regards,

Stefan

[trueangle]

@StefanOltmann! Absolutely, that was kinda unexpected. Thanks a lot for sharing it with the community.

I'm interested in ways to further improve the performance of the Lambda function. For example, is Amazon Linux 2023 faster than Amazon Linux 2?

Should be. They introduced some optimisations that could speed up the env potentially. However, the choice of the arm architecture and right memory configuration of the lambda function might have more impact on the perf.

In the discussion you mentioned cold/warm start as 350/130. 130ms still seems quite huge number for the warm start. I guess it could be optimised even further. Have you tried allocating more memory to the lambda? E.g. 512mb

[StefanOltmann]

Yes, I'm running my Lambda with 512 MB of memory. I'm not sure about the cold start time alone — the durations I mentioned are the total execution times reported by CloudWatch. When the Lambda is cold (e.g., if I haven't interacted with Alexa for an hour), it takes about 350 ms to start and handle the request. For subsequent (warm) invocations, it drops to around 130 ms.

I'm planning to try the ARM version next, and I'll also compare it with Amazon Linux 2023. I will report here.

Coming from a Java Lambda that used to take 3 seconds, I'm already quite happy with the current performance. :)

[StefanOltmann]

Quick update: Indeed changing the runtime to arm64 is well worth it.

That numbers improved a bit. I'm satisfied. :)

From a cold start:

Duration: 294.23 ms
Billed Duration: 373 ms
Memory Size: 512 MB Max
Memory Used: 44 MB
Init Duration: 78.05 ms

From a warm start:

Duration: 134.87 ms
Billed Duration: 135 ms
Memory Size: 512 MB Max
Memory Used: 45 MB

[trueangle]

Nice! Any ideas why the warm start duration takes 134.87ms?

[StefanOltmann]

No, unfortunately not. Can I somehow turn on additional analytics to get an idea?

[trueangle]

No, unfortunately not. Can I somehow turn on additional analytics to get an idea?

There's no analytics. I guess one of the options is to define own benchmarks in the code to measure performance impact