For awhile now, we offered a Matrix instance and even publicly displayed it on our site for users to register and use. While it functioned and did it's job, it was a pain with upkeep and had severe problems with the way it functioned. I have deemed it no longer important enough to work with, and decided to shut it down permanently.
My decision was made with multiple factors in mind:
Stability While the Matrix server could run fine without being touched, it ended up being difficult to deal with errors in it's functionality. There were servers we couldn't even interact with even though they were up-to spec because of errors in how the server handled their connections. There was no way to override this or clear it's cache either, making this effectively a permanent issue.
Support I have a little rule for running anything on the servers, and the most supported server for the Matrix protocol ends up breaking one of those rules. Synapse is made in Python, and I refuse to run anything using Python or NodeJS on the servers for both security and efficiency concerns. We had to settle for a lesser-supported but maintained server called Conduit, which ended up being the biggest pile of junk I've ever worked with. Not only was it a pain to update, but it ate up system resources and ran like crap. Considering the little diversity and choices within the Matrix server space, it's not a worthy path to journey. There is
Construct, which looks relatively promising considering the other projects I have explored.
Efficiency Despite being an open-source spec, neither the spec or any software seems to attempt to be efficient and effective at the same time. The spec is riddled with issues, and Matrix is still far from being decentralized. It becomes difficult to deal with a running service like this that hogs up resources to deal with a few messages. Matrix also took up 1st place in the most requests and responses received through both client communication and federations. It is inefficient for bandwidth and memory.
While I can see this upsetting some users, this decision is final and I am content knowing I will not have to touch Matrix ever again. I am working on an alternative that is an extension to XMPP, a similar yet more mature product that Matrix promised to be and wasn't. We will be extending XMPP to provide our own account system, features, and interfaces. This will be compatible with existing XMPP clients, although they won't (officially) support Worlio extensions.