17 comments

[ 3.1 ms ] story [ 46.7 ms ] thread
I built something similar 2.5 years ago in case you're interested: https://github.com/tobilg/lsh
+ dgsh [0], I wonder if one could build a serverless sh map-reduce?

[0] https://news.ycombinator.com/item?id=21700014

If you count someone else's server as "serverless", then yes: https://adamdrake.com/command-line-tools-can-be-235x-faster-...
>If you count someone else's server as "serverless"

That's literally what serverless is. Serverless doesn't mean there are no servers. It is simply the ability to run your code without having to think about the server it is running on. More server agnostic than serverless.

Very interesting work, thanks for the link! We actually missed it among the references in our paper [1] and will correct this in the journal version. Regarding the differences, lsh has an interactive mode but not sshell. On the other hand, sshell compile to native (to support massively parallel calls), and IPCs between such calls (thanks to the DSO layer).

[1] https://drive.google.com/file/d/1D7h0hoMep0W73XV_EdXPSEWUxpT...

I think the documentation should give the precise AWS permissions so that a role can be issued. Currently it only reads "# the role must have access to AWS Lambda".
This is now detailed.
Oops... I read this as "The Serverless hell", not joking.
“In the middle of the journey of our life I found myself within a dark woods where the straight way was lost.” – Dante Alighieri, Infern.io
More upvotes here ^

Edit: And TIL infern.io is a thing too

what's that? I thought it'd be related with inferno OS.
(comment deleted)
(comment deleted)
Cool idea, but wouldn't applications of this be pretty significantly limited by lambda execution timeouts?

The "process big data in the cloud" use case in the README seems like it's not really all that tenable, depending, I guess, on the structure of the data and the necessary processing steps.

sshell inherits the limitations of the FaaS infrastructure, including the time limit. Regarding the big data use case, more work is necessary to assess this, but the applications we have written so far were competing both in terms of efficiency and pricing (Section 5 in [1]).

[1] https://drive.google.com/file/d/1D7h0hoMep0W73XV_EdXPSEWUxpT...