Blog — page 2 of 38
Bioinformatics Pipelines on microVMs: Reproducibility, PHI and the 400-Tool Dependency Graph
Two stages of the same pipeline want incompatible htslib builds, the postdoc's R script needs to write into the site library, and the whole thing is bit-for-bit reproducible until somebody upgrades the host kernel. Containers fixed the packaging. The VM boundary fixes the rest.
DAMON: Measuring What a Sandbox Actually Touches
Charging a host for every guest's configured RAM is what caps your density, and it is wrong by a lot — a guest handed 4 GiB to run an idle Node server touches a fraction of it. DAMON is the kernel subsystem that can tell you which fraction, at a cost bounded by region count instead of memory size. Here is how it works, and the three different wrong answers you can get instead.
Per-Sandbox Disk I/O Throttling and the Neighbour You Cannot See
CPU contention is visible and roughly fair. Memory contention has pressure signals. Disk contention shows up as latency in somebody else's unrelated workload, with no attribution and no ticket. Here are the three places you can actually throttle it, what each one really controls, and the two layers that quietly lie to you.
Patching the Host Kernel Under Running MicroVMs
The host kernel is the thing your isolation boundary is made of. It cannot be restarted without taking every guest on the box with it, and the clock on the patch is a security clock.
PandaStack vs Depot: Build Acceleration Is Not a Sandbox API
If your Docker builds are slow and your Actions cache is a liability, buy Depot — a sandbox API will not help. But both companies bet that producing a warm environment is the product, and the difference between a layer cache and a snapshot is the most useful thing in this comparison.
PandaStack vs Namespace Cloud: Two Kinds of "Ephemeral"
Three different things are called "namespace" and only one of them is a company. Here is the disambiguation, and then the honest comparison: a managed-runner-and-cache product against a microVM sandbox API, including the two places the choice is genuinely real.
Top 8 Ephemeral Development Environment Platforms (2026)
Feature grids do not decide this. Two numbers do: how fast a fresh environment can exist, and what happens to it when nobody is looking. Eight platforms graded on both, with the substrate each one actually isolates with and the catch I would want before the purchase order.
Top 7 Disposable Postgres Platforms for Developers (2026)
Three of the seven ways to get a disposable Postgres are not products you buy. Here are all seven shapes graded on time-to-DSN, isolation boundary, whether they branch real data, and the failure mode each one actually has.
Best bare metal providers for Firecracker in 2026
Firecracker needs /dev/kvm, and that one requirement disqualifies most of the cheap compute on the market. The three tiers of hosting that can actually run it, what each gets wrong, and the 60-second check to run before you sign anything.
Server-Side Rendering Someone Else's React Component
If your product server-renders components your customers wrote, you are running their code in your process, with your env vars and your database socket. node:vm is a sandbox for values, not for a module graph that can require("fs"). Here is the boundary that actually holds, and what it costs.
One MicroVM per Crawl Job: Isolating a Multi-Tenant Crawler
If customers can type a domain into a box, you do not run a crawler — you run an interpreter for the public internet, and your tenants chose the programs. Session bleed, SSRF, bad targets, and one microVM per job.
Letting Customers Run Their Own Container Image
“Bring your own container image” is a remote-code-execution endpoint with a product manager attached. The admission pipeline, the baked-RAM surprise, and why the escape surface you just inherited belongs to the kernel rather than to Docker.
A Real Red-Green Loop for Coding Agents, One MicroVM per Attempt
The single highest-signal tool you can hand a coding agent is a test runner it can actually run. Here is the two-layer architecture that makes the loop fast enough to use, the output format a model can read, and the limits the loop exists to discover you forgot.
Workload Identity for MicroVMs That Live 90 Seconds
Your sandbox boots in 179 ms, lives for a minute, and needs to write to one S3 prefix. So there is an AWS_SECRET_ACCESS_KEY somewhere, and you already know where you put it and that it was wrong. Here is what identity can actually mean for a VM with no history.
How a Control Plane Authenticates Its Host Fleet: Node Tokens, Rotation, and mTLS
Everyone writes about how users log in. Almost nobody writes about how the machine that runs your VMs proves to the control plane that it is itself — or how you rotate that credential without taking the fleet down with it.
PandaStack vs Blacksmith: Mostly, Buy Blacksmith
These are not the same product, and for most people reading a comparison of them the answer is "Blacksmith, obviously." Here is the boundary drawn precisely, the one place the two genuinely overlap, and the technical thesis they share: that the cost of a fresh environment is the product.
A Linux Desktop per Agent: Xvfb, VNC, and One MicroVM per Session
A computer-use agent driving a mouse over arbitrary websites needs a whole desktop, and the blast radius of that is a whole machine. Here is the Xvfb + x11vnc + noVNC stack, why it belongs in its own microVM, and the parts that have no GPU to save them.
Untrusted CAD: Running Customer (and Model-Generated) Geometry Scripts in a MicroVM
A product configurator, a 3D-print shop and an AI agent that emits parametric geometry all end up in the same place: executing a script that produces a solid. "We only run geometry scripts" is not a security property. It is a hope.
Customer-Authored Policy Code: Rego, CEL, and When a Rules Engine Needs a VM
Every B2B platform eventually lets customers write the rule. The honest answer is a three-rung ladder, and most teams should stop two rungs below the one I sell — a microVM create is 179 ms p50, which is three orders of magnitude slower than a CEL eval.
Top 9 Ephemeral Development Environment Platforms (2026)
Three different products are sold as "ephemeral development environments," and most comparisons rank all three in one list. Here is the split that matters, nine platforms graded inside it, and why idle cost and TTL semantics are the two line items to read first.
Top 5 Scratch Postgres Platforms for CI (2026)
Before you shop for a disposable-Postgres product: a tuned postgres:16 container plus CREATE DATABASE ... TEMPLATE beats every platform on the board for unit and most integration tests. Five options for the cases where it genuinely doesn't, including the one where my own platform is the slowest thing here.
MGLRU and MicroVM Density: Reclaim When Every Page Is Someone's Guest
The host kernel picks reclaim victims using access information already filtered through a second kernel. MGLRU gives it generations instead of two buckets — which changes the shape of your bad days, not how many guests fit on the box.
The I/O Scheduler Inside a MicroVM Is Almost Always Wrong
Your guest's block layer is sorting requests by sector number to optimise seeks on a platter that is a file on somebody else's filesystem. Why `none` is the answer, why it still merges, and why readahead is the knob that actually moves the number.
firecracker-containerd: Running OCI Images in MicroVMs, and What It Costs You
It lets you keep your images, your registry and your containerd tooling, and swap the thing at the bottom for a microVM. That is a genuinely small diff for a large change in blast radius — and it is still the wrong tool if what you wanted was a fast sandbox API.