Live Lab


By Craig Mayhew

Everyone needs a testing ground, and for me is where I test deployment of new tech. Detailed below is future tests, and results of previous ones. As usual, it's opensource.

Future: World computer
Golem currently looks like the first opportunity to try distributing a problem over a trustless and distributed cloud.

Future: Decentralized file storage
Sia currently looks like the best opportunity for a decentralised data backup. In future it may become possible to test serving files from the edge.

2020: IPNS
The IPFS deploy of is online. This uses _dnslink TXT entries and a (nearly) solid deployment pipeline in with a IPFS pinning service. The next step will be deploying WebAssembly to IPFS for, making sure it's using a config file in repo and automating the update of TXT _dnslink.

2019: Rust -> WebAssembly
Rust puts safety first, while being an incredibly fast compiled language. WebAssembly is a binary, fast language that runs in the browser and rust can compile to webassembly! Naturally I rewrote bigprimes as a single page webapp from the ground up in rust, compiled to webassembly and saw the AWS bill for reduce by 90% (all compute is now client side)! I live streamed the second half of this rewrite: Zero Point Craig: Rust -> WebAssembly

2018: IPFS is now available via IPFS. As a pregenerated static website, it was simple to serve it from IPFS as well as AWS cloud front. The biggest challenge were links to stylesheets and images, all had to be converted to relative urls depending on what page they were served on!

2017: Serverless
Moving an old php powered website to serverless in a lift and shift fashion was surprisingly simple and resulted in instant cost savings. However, the AWS api gateway service is causing costs to be higher than expected. This could be due to browsers polling the site for icon files that don't exist on every page view. More investigation is needed.

2014: AWS S3 Torrents
This was very simple and worked without any unexpected side effects

2009: Distributed computing
During the trial we didn't see any badly behaved clients, but using an SQL database did show scale limitations. A schemaless approach should be used instead.

© 2005-2024 Craig Mayhew