Post

Browser-grade sandboxing is what’s needed to run untrusted code. We run untrusted code all the time in our browser. That’s why WebAssembly is appealing. I know from experience, though, that working in standards is a difficult game when you want to move fast and be innovative. The long-game can win.

I completely miss costs in this discussion, it's not the primary driver (and should not be), but has huge implication on technical decision and business overall.

That's a good distinction in the nuances because no one has infinite time and ability and not everything needs to be specially made Working on my cars often requires special tools. It'd be nice to have them but I only need them once a year and I can get by with vice grips

Sweet sweet shared memory architecture So many underutilized macbooks

I’m seeing this already tbh

It's a big selling point that a minority of people actually need It also turns out to be mostly a US problem. Plenty of other countries have similar requirements and held on to the control of their data

Is this to be discounted as more and more compliance frameworks are necessary for enterprise sales deals? Like ... I think it is hard to sell enterprise software products today without being PCI/SOC 2 compliant at a bare minimum.

Many SaaS/PaaS providers matured a lot. When we were evaluating a service/platform, we could just go pull this list of compliance certs, which basically shifted a good bit of the liability onto the provider.

I was more thinking up the stack. I remember looking at the cloud portions of Merki right around the time of the GDRP scramble and having to think about data custody differently. Merki technically wasn't compliant with how they were storing data.

I'm less talking about ISO2701 or SSAE18 ... blah blah blah on (The physical layer stuff ... because I agree any DC worth paying this should be table stakes)

No one would suggest that tech startups start writing their own compilers because they'll be better suited to their business.

It's gotten better in recent years. StatefulSets, and CSI make it easier to pin resources. We used K8s to build test / validation HDFS clusters years ago.

Well their competitor coder realized that several years ago and came up with a very similar architecture, but I think gitpods problem is they jammed things into the app hosting portion of k8s instead of using the core api/controllers/operators instead.

Thanks you, this is really interesting to read

Yes, I had a similar experience in my previous org. Enterprises are pushing K8s everwhere but edge require specialized solutions tailored to its unique limitations and usecases.

I don't like to post things from the other app, but I did think this falls in line with what you are talking about: x.com/robin_faraj/...