,
5 tweets,
1 min read
Read on Twitter
Statelessness is not the critical property of #serverless compute, it's ephemerality. Being positively limited in duration means the provider can *transparently* manage the platform, no scheduled (or unscheduled, in Fargate's case) downtime needed.
Currently, most serverless compute is time-limited, run in response to fixed-size(-ish) events. But finite-input-limited compute is possible; this is AWS Batch's model.
Several acadamic papers, including the recent one from Berkeley, have equated serverless with existing FaaS models and complained of its inapplicability to big data processing. I think this is a failure of imagination.
I expect to see more-stateful #serverless compute models in the future for data processing and analytics use cases. I don't think they will look like those proposed by these papers, where features used by traditional compute are ported to existing FaaS models
In the same way that event-driven FaaS enabled more-managed compute than previous PaaS offerings that looked more traditional, I think we'll see different data processing architectures that'll better enable provider management of the platform without sacrificing high performance
