What Game Server Tick Rates Teach About Polling Intervals
The first time my little Minecraft server printed "Can't keep up! Is the server overloaded? Running 2034ms or 40 ticks behind," I assumed the machine was too small.

The first time my little Minecraft server printed "Can't keep up! Is the server overloaded? Running 2034ms or 40 ticks behind," I assumed the machine was too small. It was a four core box with 8 GB of RAM serving six friends, which should have been plenty. The real problem turned out to be a hopper farm someone had built near spawn, and it took me an embarrassing weekend to find it.
That warning taught me more about scheduling than any monitoring tutorial I had read. A game server lives or dies by a fixed loop that has to finish on time, and so do the cron jobs, health checks and metric scrapers that most of us run at home without thinking much about how often they fire. If you have ever had a backup script overlap with itself, or an uptime monitor that never noticed a short outage, you have met the same problem from the other side.
A tick is a promise with a deadline
A game server does not react to every keypress the moment it arrives. It collects input, advances the simulation by one step, sends the new state out, and then waits for the next step. That step is the tick. Minecraft runs at 20 ticks per second, which gives each tick a budget of 50 milliseconds. If the work inside one tick takes 70 ms, the server falls behind, and after enough of those it starts skipping ahead to catch up. Players see this as mobs teleporting and crops that seem to grow in bursts.
Competitive shooters pick much tighter numbers because their players notice the difference. Here is roughly where some well known games have landed, based on what the studios have published or what community testing has measured:
| Game | Server update rate | Time budget per step |
|---|---|---|
| Minecraft (Java) | 20 ticks per second | 50 ms |
| Fortnite | about 30 Hz | about 33 ms |
| Overwatch (after the 2016 change) | about 63 Hz | about 16 ms |
| CS:GO matchmaking | 64 tick | 15.6 ms |
| Valorant | 128 tick | 7.8 ms |
Counter-Strike 2 complicated this picture in 2023 with its "sub-tick" system, which timestamps actions between ticks rather than raising the rate. That is a useful reminder that a higher number is not the only lever. Sometimes you keep the loop the same and record events more precisely inside it.
Why your home lab has ticks too
Every scheduled thing you run is a loop with an interval and a budget, even if nobody calls it that. A cron entry of */5 * * * * is a tick rate of one per five minutes. A Prometheus scrape config is a tick rate. So is the 60 second default heartbeat in Uptime Kuma, and the OnCalendar line in a systemd timer.
The game server framing helps because it forces two questions people usually skip. What is the budget, meaning how long can one run take before it collides with the next one? And what is the cost of a missed step?
The overlap problem
My nightly rsync job to an external disk was scheduled every hour during a migration. One evening the disk got slow, a run took 74 minutes, and cron cheerfully started a second copy on top of the first. Two rsyncs writing to the same target is not a disaster, but it doubled the load and stretched both. This is the Minecraft "can't keep up" message in a form that nobody prints for you. The fix is boring and reliable: wrap the job in flock -n /tmp/backup.lock so a second run exits if the first is still going, and log when that happens. systemd timers handle this better by default, since a service that is still active will not be started again by its timer.
The blind spot problem
The other failure is sampling too slowly to see what matters. Prometheus ships with a global scrape interval of one minute, and many people never change it. If your reverse proxy falls over for 40 seconds between two scrapes, the graph shows nothing at all. A shooter running at 20 Hz has the same issue with fast movement: a player can cross a doorway between two updates and the hit registration has to guess.
Faster is not automatically better
The common advice in self hosting forums is to tighten everything. Scrape every 5 seconds, ping every 10, run the sync every minute. I followed that for a while and regretted it.
Game studios do not pick 128 tick for every mode, and there is a reason. Valve kept CS:GO matchmaking at 64 tick for years largely because doubling the rate roughly doubles the CPU cost per match, and that cost is multiplied by every server in every region. At home the multiplication is smaller, but it is real. When I dropped my Prometheus scrape interval to 5 seconds across 14 targets, my small TSDB on an SD card grew fast enough that I had to cut retention from 30 days to 9. The graphs were prettier. I did not learn anything new from them.
Pick the interval from the shortest event you actually need to catch, then stop. Everything faster is paid for and rarely looked at.
For most home services, a short outage that nobody noticed was not worth detecting. The outage you do care about, the one that lasts until morning, a one minute check catches easily. Where speed matters is the other direction: alerting. A check every 30 seconds with an alert that waits for three failures in a row gives you a 90 second window that filters out blips without hiding real problems. Uptime Kuma calls this "retries" and it is the single setting I would change first.
Profiling a slow loop, the game server way
When a Paper or Purpur Minecraft server lags, the community has a well worn routine. You run /mspt to see milliseconds per tick, then use the spark profiler to find which chunk, entity or plugin is eating the budget. Nine times out of ten it is one thing: a redstone clock, a pile of dropped items, a mob farm with 400 chickens. You fix that one thing and the server is fine again.
The same habit works for scheduled jobs. Before you buy a bigger box, time the job. A simple time ./backup.sh in the log, or the duration fields that systemctl status already shows, will tell you whether the job fits inside its interval with room to spare. My rule of thumb, borrowed from that 50 ms Minecraft budget, is that a job should finish in under half its interval on a normal day. That leaves headroom for the slow day when a disk is busy or the network is congested.
If you want the bigger picture of how small servers stay responsive, I wrote about the network side in why low latency matters so much to players, and the hardware side in running a community server on a small budget.
What I would change this week
Open your crontab and your systemd timers side by side and write down, for each job, the interval and the longest run time you have seen in the logs. If you do not have the second number, that is the first thing to fix: add a line that logs start and end times. Any job whose longest run is more than half its interval gets a lock or a longer interval. It takes about twenty minutes for a typical home lab, and it is the closest thing a self hoster has to running /mspt on the whole house. More server stories live in the games section.
More in Games
Games
How Online Draw Games Keep Results Fair at Scale
Online draw games feel like the simplest thing on the internet.
Games
How Online Game Webhooks Explain Automating Alerts for Your Own Server
If you have ever set a game to post your match results into a group chat, you have already used a webhook.
Games
How Online Game Lobbies Explain Load Balancing
When you join an online game you rarely think about which machine is actually hosting your match.