One dataset. Five machines. Your code on every node, and every line it writes streaming back to a single page as it happens.
GK Grid takes a shared database, pushes it out to four worker machines, runs the processing you wrote on each of them, and collects the answers back into one place. The whole run is visible while it happens — not as a progress bar, but as the real log line each machine is producing at that moment.
Every part of the platform is plain PHP and plain SQL. There is no scheduler to learn, no cluster manager to install, and no configuration language between an idea and a running job.
A browser is the whole client. Drag files onto a page and the run starts. No agents on the nodes, no command line, no build step.
One panel per machine, filling as work happens, each with its own running clock. You watch the slow node while it is still slow.
The number of packages you send is the number of nodes that run. One worker or four, decided at the moment you press the button.
Nine groups can work at once. Directories, databases and log streams are kept apart, so nobody sees or breaks anyone else.
A system view opens all five machines side by side: files, tables, row counts and contents, without touching the servers.
One file runs on the node and you write it. Use the supplied transfer helpers, replace them, or work an entirely different way.
Ordering is enforced by the platform: the primary always completes before a single worker begins, so shared preparation is finished before anything depends on it.
The entire platform is a handful of files. Nothing is hidden in a framework, which is the point: a student can follow every hop a byte makes.
| Runtime | PHP 8, PDO, MySQL 8 |
|---|---|
| Front end | HTML, CSS, a little vanilla JavaScript |
| Dependencies | None. No framework, no package manager, no build |
| Transport | HTTP between nodes, gzip-compressed SQL payloads |
| Extraction | In-process, never a shell command |
| Isolation | One directory, one database and one log stream per group |
| Recovery | Every node reports completion or failure; a dead node cannot hang the run |
| Interface | Desktop and mobile, one page |
Distributed computing stops being abstract the moment you can watch four machines think at once.