slow performance

[lukaspour]

Hi, I am seeing quite alow performance in resizing pictures, is it normal?

Resizing FullHD image to 2 sizes takes almost 7 seconds. Is this a normal behaviour? Thank you.

[ysugimoto]

It seems to be a slower performance.
Can you show me your configuration and image you have processed?

[kdybicz]

In my work-related project I'm converting source file to 7 differently cropped and resized files and it always takes less that 2 sec. I believe it's due to the memory setting, which in my case is 1280MB. If you are afraid of the higher costs keep in mind that image processing is memory and CPU consuming process, so the waker environment you will use, the longer processing will take. As Lambda charges you per each 100 ms then it's worth trying to find compromise in memory settings. For me it was 1280 MB.

[lukaspour]

@kdybicz I have tried to change the memory settings but without any difference - logs always show memory usage around 200MB so how would 1280M speed it up?

Here is a newest example:

The picture:

The ZIP file with config:
DevAvatarsResize-f3002945-490e-4827-9ca3-127830104060.zip

@ysugimoto thank you for helping us, we just can not find any reason why it is slow.

I believe the size of the image corresponds to the transformation time 1MB takes approx 1-1.5sec to resize.

[kdybicz]

I did run some test with provided config.json and image, and I must say that you are right. Processing images is resource and time consuming. In this case looks like it's all about time, because there are some resources left, that are not used. We can try to introduce concurrent image processing in same Lambda instance or distributing those tasks to different instances. Either way it's not easy task and could take some time to prepare proper changes. Maybe you would like to help us with preparing such new feature?

[lukaspour]

@kdybicz Thank your for testing it. Unfortunately I am not a programmer so I do not know how to help but I am happy to help you with testing.

[kdybicz]

@lukaspour In that case I will leave this ticket open, just to be in touch in case we managed to solve this issue. Also, maybe someone else will experience same problems and will be able to provide fix proposal.

PS. I'm working on introducing SNS support right now. I believe it could be used as kind of workaround in your case. You will just need to set few separate functions, each for different output format, and subscribe them for the same SNS topic. I'm not sure yet if this is 100% doable, will let you know when SNS support will be in place.

[schmoove]

This could be a good case for switching from ImageMagick to Sharp:

https://github.com/lovell/sharp

Generally 4-5x performance, but a lengthier & more involved installation process:

http://sharp.dimens.io/en/stable/install/#aws-lambda