Brendan and the Candi team mentioned they were getting less than 10ms response times from dynamodb with Golang. This seemed rather too-good to be true, since we'd been seeing a typical response time around 60 milliseconds with the nodeJS aws-sdk.

So, I decided to do a quick roundup in a few different implmentations to see if I could answer:

• Were their results accurate? Is the AWS-SDK for node that much more terrible that it's roughly 6 times slower?
• if so, Is this becuase dynamic/static typing?
• Is this simply the different consistency models being defaulted to different options (ie eventual vs strong)?
• Could this just be a maturity thing/some weird thing I've not thought of?

So:

1. I turned off strong consistency for all implementations (where it was obvious, dunno about amazonica/faraday, I think it defaults here anyway).
2. Did a simple loop for a record, waiting for 1 second between calls so as not to make it an IOPS bound thing
3. Threw the applications into an ec2 instance and executed them
4. Recorded results

raw results are just a stdout dump in /results/*.log

1. Yep, NodeJS appears to be performing terribly.
2. This can't be a dynamic-languages-aren't-fast thing, python's only a couple of milliseconds off golang
3. I still don't know what's going on with node. I'll have to do some flame-graphs or something.
4. Clojure/JVM seem to take a while to 'warm up', then they perform reasonably well