blog
Blog — page 19 of 39
Posts 433–456 of 928, newest first.
Runnable Code Playgrounds in Your Docs, Safely
A Run button in your docs is a free, unauthenticated compute API wearing a documentation costume. Here's the threat model, the latency budget, and the per-run microVM architecture that survives launch week.
Ajay KumarFirecracker on arm64 vs x86_64: What Actually Differs
Same API, same virtio-over-MMIO, same jailer. Different boot protocol, different interrupt controller, different clock — and snapshots that never cross the architecture line. Here's the honest diff.
Ajay KumarConfidential computing for microVMs: SEV-SNP, TDX, and arm CCA explained
Ordinary virtualization protects the host from the guest. Confidential computing tries to protect the guest from the host — including from the operator holding the keys to the datacenter. Here's what the hardware actually does, what it very much does not do, and why you probably don't need it.
Ajay KumarLinux capabilities, explained for sandboxing: five sets, one bounding ceiling, and the bits that are still root
Capabilities split root into about forty bits so you can hand out one instead of all of them. Worth doing, cheap, real — and about a dozen of those bits are still root wearing a lanyard.
Ajay Kumario_uring and sandbox security: the fast path that walks around your syscall filter
We invented an interface that lets you do I/O without making syscalls, and then acted surprised when the syscall filter stopped seeing the I/O. io_uring is a genuine performance win and a genuinely awkward sandbox problem — here's the mechanism, the mitigations, and why a guest kernel changes the math.
Ajay KumarBest Jupyter notebook hosting platforms in 2026
JupyterHub, Colab, Deepnote, Hex, Databricks, SageMaker Studio, Modal, Binder, Codespaces, and PandaStack — judged on who runs the kernel, what it costs while idle, and whether you can safely run somebody else's notebook.
Ajay KumarModal vs RunPod: Serverless Python or Raw GPU?
Modal sells you a programming model that happens to have GPUs behind it. RunPod sells you GPUs that happen to have a programming model in front of them. Which sentence describes your team decides this, and almost nothing else does.
Ajay KumarIP Address Planning for a MicroVM Fleet
Subnet exhaustion is the outage that gives you no warning and then all of it at once. Here's the address math for a microVM fleet, the leak that drained our pool, and the durable allocator that fixed it.
Ajay KumarThe best Google Cloud Run alternatives in 2026
Nobody leaves Cloud Run because it is bad. They leave because their workload stopped being a request/response container — and Cloud Run's whole model is built around that shape.
Ajay KumarThe best Gitpod alternatives in 2026
Gitpod taught everyone to expect a dev environment from a URL. Picking a replacement means deciding which half of that promise you actually needed: the ephemerality, or the machine.
Ajay KumarThe best Amazon RDS alternatives in 2026
RDS is the safe choice, and safe choices get expensive quietly. The alternatives are worth a look — but only if you can name which RDS property you are willing to give up.
Ajay KumarThe best Coolify alternatives in 2026
Coolify is a genuinely good deal until the single VPS running everything becomes the thing you worry about. Here's what to move to, and when moving is the wrong call.
Ajay KumarThe best Laravel hosting platforms in 2026
A Laravel app is never just a web process. Queues, the scheduler, storage, and sessions decide which host suits you far more than the price per month does.
Ajay KumarThe best Bun hosting platforms in 2026
Half the platforms that advertise Bun support use it as a package manager and then run your app on Node. If your code calls Bun.serve, that distinction is your entire deployment.
Ajay KumarThe best React hosting platforms in 2026
A React SPA is a folder of static files. Everything expensive about hosting one comes from the three things that folder cannot do by itself.
Ajay KumarHow to deploy a SvelteKit app without Docker
SvelteKit's adapter system means your deployment is decided before you ever pick a host. Get the adapter right and the rest is a Node process on a port.
Ajay KumarHow to stream command output from a sandbox
A ten-minute build behind a blocking exec call looks identical to a hung one. Streaming is not a nicety; it is how you tell the difference.
Ajay KumarHow to persist data in a sandbox
Everything you write inside a sandbox disappears when it does. That is the feature. The skill is knowing which of four escape hatches fits the data you actually need to keep.
Ajay KumarBuilding a data clean room on microVMs
An advertiser and a publisher both want the overlap and neither will hand over their rows. A clean room is the machine that computes the answer and then stops existing — which is a much easier thing to build on a microVM than on a promise.
Ajay KumarRunning academic artifact evaluation on microVMs
An artifact-evaluation reviewer gets a tarball, an install.sh with sudo in it, and two weeks to decide whether a paper's numbers survive contact with a second machine. A snapshotted microVM turns that into a machine you can hand to the next reviewer.
Ajay KumarHermetic Builds and SLSA Provenance on MicroVMs
If the signing key sits on the build runner, every postinstall script in your dependency tree is one curl away from signing whatever it likes. Put the signer outside the guest.
Ajay KumarPer-Partner Isolation for B2B EDI and File Integrations
One trading partner changes a field width without telling anyone, their 900MB interchange OOMs the shared worker, and every other partner's overnight run dies with it. Give each partner a machine of their own.
Ajay KumarFirecracker Guest DNS Resolution Explained
"It's always DNS" is a cliché precisely because it keeps being true. Inside a Firecracker microVM there is no DHCP server and no resolver — name resolution is entirely something you wire up, bake in, or forget to.
Ajay KumarCPU Steal Time in microVMs: Your VM Isn't Slow, It's Waiting
The CPU wasn't slow, it was in a meeting. Steal time is the one number that tells you whether a tenant's slowness is their code, their storage, or your overcommit — here's the mechanism, the misreads, and the fix order.
Ajay Kumar