3 ms·
The posters here who are commenting on what should be (often largely what Kubernetes already is) would do well to read the paper about the Omega scheduler. htt
by ted_dunning 5y ago
The posters here who are commenting on what should be (often largely what Kubernetes already is) would do well to read the paper about the Omega scheduler.
https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/41684.pdf https://static.googleusercontent.com/media/research.google.c...
It is almost the rule that none of us has the breadth of experience to understand all of the problems that need to be solved. In fact, an explicit goal of Kubernetes is that almost all users should not need to understand much of how the system actually works.
Frankly, most users have neither the time, inclination or intellectual capacity to understand more than a tiny part of the systems they work on. That includes most people who say they do understand those systems completely ... a few short questions can generally expose how little they know about important aspects.
That ignorance is a measure of success, btw, not failure.
- simonw 5y agoThe problem here is the law of leaky abstractions: https://www.joelonsoftware.com/2002/11/11/the-law-of-leaky-abstractions/ https://www.joelonsoftware.com/2002/11/11/the-law-of-leaky-a... Systems that abstract away the complexity for you work fine up to a point... but inevitably you'll end up debugging something caused by a lower layer than the one you are familiar with, at which point you need to dig in and figure out what that abstraction is doing for you. It's not possible to understand everything, but the more you understand the more productive and effective you'll be. I see my overall career partly as an ongoing quest to deepen my understanding of the various abstractions I rely on.
- bogomipz 5y agoWasn't Omega the successor to Borg? I'd be curious what's the status of it today at Google?
- dekhn 5y agoI believe there are several public articles what happened. Some omega features were folded back into borg, omega development pivoted to k8s development. Most of google still runs on borg/internal but more and more people are moving projects to GCP-hosted cloud. That only really applies for things that can run Outside, and for which k8s is "good enough". Borg will continue to exist, roughly in its currently for, for at least 20 more years if Google doesn't get shut down by DOJ. It would simply not be cost effective to migrate large parts of GOogle's infra off Google borg/internal. In my experience, absolutely nothing in the world matches the experience inside of google; enormous number of enormous clusters and gobs of networking within and between them, plus a heavy set of existing workloads constantly testing it, all while people constantly deploy new versions of borg, filesystems, database and applications from a single codebase. Just as importantly, there was a massive culture of cluster nerds (SRE) that ran the clusters very efficiently, but also with care and attention to a wide range of users, including ones that weren't on Big Teams. I never, ever would have been able to build and run exacycle, an idle cycle harvester that provided at least 4 significant contributions to science.