Hacker News Reader

Comment by randoramax | Hacker News Reader

We need to think of ways to run the various Matrix, Mastodon, peertube etc in tiny boxes in our homes! There is no real reason for not promoting self hosting. The same way i buy a small Nas or an amazon firetv stick, with apps, I would buy a small box with pihole, mastodon and matrix.

View on HN

Have you tried running Mastodon and Matrix on a Pi? I'm not so confident it'd work, these aren't lean servers.

I would love a tiny server I can run on my home server (for my tiny message volume), but Mastodon and Synapse are optimized for large servers.

synapse is actually quite tiny, but its use of an sqlite db on disk and other disk-write related activities, over a long term of months or years, seems quite likely to destroy a raspberry pi microsd card in much the same way that ordinary logging will destroy a microsd card. yes the rpi4 can boot from a USB-connected 'real' sata device now, which is an improvement.

> synapse is actually quite tiny

Hmm, is it really? My server has a SSD so I don't care about writes, I just want it to be low on CPU and memory requirements.

an entire synapse with its python environment is about 150MB of disk space occupied, before any user data starts getting written to the database.

chat for a group of about twenty people runs fine on a VM with 512MB of RAM. After boot about 300MB of the system RAM is occupied by the base minimal debian OS and other daemons (snmpd, etc), leaving plenty for python3.

this same machine also has a minimal nginx setup as the standard reverse proxy configuration for TLS1.2/TLS1.3 in front of the synapse daemon, which isn't set up for any crypto and only listens on localhost.

It's worth noting that this sounds to be an unfederated server.

The reason Synapse has a reputation for being resource hungry is on publicly federated servers: if one of the users starts going and joining a bunch of large busy public rooms from their new server, then those rooms get dutifully replicated onto the server, which inevitably consumes resources.

The more users and the more servers in the room, the more changes of gnarly forks and merge resolution problems, and the more the risk of CPU spikes. We're constantly working on the memory footprint and state resolution algorithm (e.g. https://github.com/matrix-org/synapse/blob/develop/docs/auth... landed a few days ago), so the situation is improving, but this is the root cause.

Yes, absolutely, this is for company internal use only. I imagine the resources for a server that talks to the outside world are much greater.

I have tried it on a RPi3 and it is painfully slow. I would not recommend this to anyone.

There's been work on integrating the server and client together in a p2p fashion.

Then you are essentially self-hosting on your phone/laptop when you run the client.

one of the challenges with that from the perspective of an ISP, is that residential broadband connections are very often highly asymmetric in bandwidth.

for example on two of north america's largest DOCSIS3/DOCSIS3.1 based cable modem operators, you can get 200-600 Mbps downstream speeds, but uploads may max out at 16-18 Mbps. Depending on your exact location. This is because of how RF channels are bonded together by the CMTS operator.

lots of other access technologies will face significant challenges and capacity constraints if a sizeable percentage of residential broadband end users start trying to actually use their upstream bandwidth at a greater rate than they do now, when averaged over the traffic for hundreds or thousands of individual end point users.

Reply on news.ycombinator.com