1. 6

  2. 3

    I wonder how well 64mb instances will work in practice - box packing that into machines is likely to cause a huge number of processes (not a major issue - I have boxes with 50k+ threads on them) but I’ve found a major issue is crazy context switching which hits performance in a big way.

    Personally speaking; If I was building a business, I would pick one of the established players (AWS, Google, Joyent). Picking a niche player seems like a recipe for being forced into a later migration.

    There are of course interesting features, but I think YAGNI often applies and these need to be offset against the added risks.

    1. 2

      Personally speaking; If I was building a business, I would pick one of the established players (AWS, Google, Joyent).

      I agree, though it’s also priced at about ½ Joyent’s prices when comparing same-RAM-allocation containers (e.g. 1.44¢/hr vs. 2.6¢/hr for a 1-GB container), so it seems more like it’s competing in the “budget cloud” category, which has somewhat different criteria and targets a different set of customers. Mostly just has to be cheap-ish and good enough. Whether it fits that niche depends on how much performance lags though… if it’s ½ the price but also ½ as fast, then it gets a bit less interesting, unless you’re purely RAM-constrained.

      1. 1

        That’s Digital Ocean’s game, right? How does it compare to them?

        1. 2

          It’s roughly the same price as DigitalOcean comparing equal RAM sizes. I suppose you’d only go for Hyper_ if you found their container-submission engine more in line with your needs than DO’s VM offering, otherwise DO is obviously more established.

    2. 2

      Has anyone used them before? Are there any published benchmarks?

      1. 7

        I haven’t, but I do recall arguing with one of their staffers on HackerNews a while ago.

        The marketing copy on the site still seems to make out like it’s some amazing solution for running containers without virtual machines, but as far as I can tell it’s still really just a “one container per VM” model. Rather than require you to spin up a VM and put containers in it, they will take your container request and wrap it in its own VM on their platform for you. It would appear that most (or all?) of the perils of hardware virtualisation (like having two OSes between you and the hardware, limiting the lower bound on latency and the upper bound on performance) still apply.

        By way of full disclosure, and to provide context for my disagreement, I work at Joyent. Our container thing, Triton, actually does run containers directly on the metal without another layer of hardware virtualisation. As an added bonus, we implement the Docker remote API, so you can just use the real docker CLI rather than something custom (like the hyper CLI they talk about).

      2. 1

        The logo appears very similar to Heroku or is it just me?