Thanks! I’ll admit, hypriot is making it all a lot easier. Before I’d started with wheezy, then updated to jessie (to get a bare minimum version of Docker), then compiled Docker 1.4.1, then was working on bootstrapping a RPI golang in order to compile swarm…. Much, much, much easier now!
Wouldn’t one Intel NUC running CentOS and a bunch of Docker containers (or qemu-kvm VMs) get you much more compute with smaller weight/power/footprint a whole lot easier ?
(and yes, I get the ‘fun’ part of building a cluster of $50 systems if the goal is experimentation)
I’d have to see some benchmarks before I’d say so for sure…. I’ve done some basic comparisons between the Pi’s and desktop systems, and truthfully it really depends on the type of work being performed as to which is more effective in cost/power. I’ve not investigated using the GPU yet; I expect for certain types of problems they would be even more cost effective.
Admittedly, I started doing this in part for fun — I wanted to have a easily transportable cloud. However, the more I’m getting into it, the more I’m realizing that I can do some really useful work for low cost. For instance, I attached a usb volt meter yesterday and a Pi B+ which was doing some fairly intensive calculations in a jvm (calculating Pi with a very inefficient algorithm) was using 1.26 watts.
Once I get a bit further along, I plan to do some comparisons — I have quad and 8 core boxes at home, too…
One of the areas in which I’m interested is in on-demand scaling and scheduling. It’s possible to do it with a single box, but harder… Capacity modelling, too is an interest. So…. long and short is that I’m scratching several itches. (and having fun, too!)
4 comments
Skip to comment form ↓
stefanscherer
March 4, 2015 at 5:25 pm (UTC -5) Link to this comment
Wow! That looks promising!
Matt Williams
March 4, 2015 at 10:08 pm (UTC -5) Link to this comment
Thanks! I’ll admit, hypriot is making it all a lot easier. Before I’d started with wheezy, then updated to jessie (to get a bare minimum version of Docker), then compiled Docker 1.4.1, then was working on bootstrapping a RPI golang in order to compile swarm…. Much, much, much easier now!
Thanks again!
Vince Skahan
March 30, 2015 at 4:02 pm (UTC -5) Link to this comment
Wouldn’t one Intel NUC running CentOS and a bunch of Docker containers (or qemu-kvm VMs) get you much more compute with smaller weight/power/footprint a whole lot easier ?
(and yes, I get the ‘fun’ part of building a cluster of $50 systems if the goal is experimentation)
Matt Williams
March 30, 2015 at 5:45 pm (UTC -5) Link to this comment
I’d have to see some benchmarks before I’d say so for sure…. I’ve done some basic comparisons between the Pi’s and desktop systems, and truthfully it really depends on the type of work being performed as to which is more effective in cost/power. I’ve not investigated using the GPU yet; I expect for certain types of problems they would be even more cost effective.
Admittedly, I started doing this in part for fun — I wanted to have a easily transportable cloud. However, the more I’m getting into it, the more I’m realizing that I can do some really useful work for low cost. For instance, I attached a usb volt meter yesterday and a Pi B+ which was doing some fairly intensive calculations in a jvm (calculating Pi with a very inefficient algorithm) was using 1.26 watts.
Once I get a bit further along, I plan to do some comparisons — I have quad and 8 core boxes at home, too…
One of the areas in which I’m interested is in on-demand scaling and scheduling. It’s possible to do it with a single box, but harder… Capacity modelling, too is an interest. So…. long and short is that I’m scratching several itches. (and having fun, too!)