The apocalypse is nigh. Soon, binary executablesand containers in object stores will join the manyWeb-based pipelines and the several virtual machine imageson the dystopic wasteland of "reproducible science."
Anyway.
I had a conversation a few weeks back with a senior colleague aboutcontainer-based approaches (like Docker) wherein they advocated theshipping around of executable binary blobs with APIs. I pointed outthat blobs of executable code were not a good replacement forunderstanding or a path to progress (see my blog post on that) and theyvehemently disagreed, saying that they felt it was an irrelevantpoint to the progress of science.
That made me sad.
One of the things I really like about Docker is that the communityemphasizes devops-style clean installs and configurations overdeployment and distribution of binary blobs (images, VMs, etc.) Let’smake sure to keep that; I think it’s important for scientific progressto be able to remix software.
I’ll just close with this comment:
The issue of whether I can use your algorithm is largely orthogonalto the issue of whether I can understand your algorithm. The formeris engineering progress; the latter is scientific progress.
–titus
p.s. While I do like to pick on the Shock/Awe/MG-RAST folk becausetheir pipeline is utterly un-reusable by anyone, anywhere, ever, I ambeing extremely unfair in linking to their paper as part of this blogpost. They’re doing something neat that I am afraid will ultimatelylead in bad directions, but they’re not espousing a binary-only viewof the world at all. I’m just being funny. Honest.
p.p.s. Bill Howe and I also agree. So I’m being multiply unfair. I know.
