First tests of our distributed storage on the new hardware, a single VM disk hit:
-> 250K 4K random-read IOPS
-> 180K 4K random-write IOPS
Who else can do that?
AWS io2 can. A maxed-out volume costs roughly $10K/month. Storage only.
First benchmarks of the new ServersCamp hardware.
Geekbench 6 single-core: 3458
PassMark single-thread: 4765
For context: Hetzner's dedicated vCPU class scores 1876. Shared - 2083 at best.
I knew my SSH was being brute-forced. Everyone knows that, in theory.
Then I built a dashboard that shows it live - IPs, countries, ports, attempt counts, reputation flags. Watching it happen to my own box felt different.
And once you see it, you kind of want it gone.
Todayโs โI knew this, but stillโ moment: I pointed a test load balancer at an empty server. No website, no domain, no shared IP.
In 24h, 113 sources from 20 countries found it and started probing.
Public IPs are never really idle.
https://t.co/9uCZO3d7kR
On January 1st my cloud got its first public tester.
Since then, most of my attention has gone into stability and new features.
Recently I turned on test billing to debug how the platform meters resources, and built this dashboard to make the numbers easier to inspect.
It is not confirmed MRR yet, but seeing the meter move makes me happy.
Alright. Backups and snapshots are done. Probably the most brain-melting thing I've built. I didn't want to clone anyone, so from the start I did it the way I see it:
1. Snapshots are separate disk copies, in the same SDS as the original disk. Fast, hot, recovery level is operational, saves you from your own mistakes.
2. Backups. This is where I dragged in enterprise and their compliance stuff. Backups are encrypted, backups are compressed, by default they go to another country, another storage, another organization. Their failure domain is all of ServersCamp, or a whole country. They get regularly tested for restore: checksum verification, plus a check that boot actually works. Also separate GFS presets.
What I like most is how easy it turned out for the user. No settings needed, just one checkbox and you get an enterprise DR backup.
@Shpigford Working on exactly this stack. Postgres + agent-native MCP provisioning shipped, Redis coming. Not ready to recommend yet, happy to ping you when itโs there.
@_swanand Don't blame Google too fast, they most likely did what they were supposed to. Policy violation, unpaid invoice, abuse report... accounts don't just get blocked for fun.