Your hosting panel shows pretty graphs: CPU, RAM, disk, network. Something spikes at 2 AM. Is that a problem or Tuesday?
Graphs without context create panic upgrades and ignored real issues. Here is how we read resource charts on Canadian VPS and managed servers, and what common spike shapes actually mean.
CPU: bursts vs sustained pegging
Short spikes to 80-100% during backup windows, cron runs, or cache warmups are often normal. Watch duration. Thirty seconds is different from thirty minutes.
Sustained high CPU with slow site performance usually means PHP workers, MySQL queries, or a runaway process. Check top processes, slow query log, and whether traffic actually increased.
Low CPU but slow site can mean waiting on disk I/O or external APIs, not lack of cores. Do not upgrade CPU until you know what is blocking.
On virtual servers, look for CPU steal time if your panel exposes it. High steal means the hypervisor is oversubscribed and your VM is waiting for physical CPU. That feels like "we need more cores" but might mean noisy neighbors or host contention. Talk to the host.
Load average: not the same as CPU percent
Load average counts processes waiting for CPU and disk. A load of 4 on a 4-core box can be fine under burst. A load of 12 on a 2-core box means a line of unhappy tasks.
Compare load to core count. Rising load with flat traffic often points to I/O or lock contention, not viral traffic.
RAM: used vs cached vs swap
Linux uses free RAM for file cache. "90% memory used" in a naive graph might include cache that releases under pressure. Look for swap usage climbing. Swap activity means the server ran out of comfortable RAM and started paging to disk. Performance tanks.
WordPress sites with large object caches, PHP-FPM pools sized too high, or MySQL buffer pools set aggressively can eat RAM without high traffic. Spikes after deploys may mean a new plugin or mis-tuned pool.
OOM kills (processes disappearing from logs) mean you genuinely need more RAM or fewer hungry workers.
Disk space vs disk I/O
Disk full graphs are obvious until they are not (inodes full with "space available"). Separate metric.
Disk I/O wait (iowait) spikes during backups, large imports, or database maintenance. If iowait stays high during normal browsing, check slow queries, missing indexes, or storage on overloaded shared nodes.
SSD vs HDD matters for database-heavy apps. High IOPS workloads on budget storage show up as latency, not always as CPU.
Network: bandwidth vs packets
A sudden outbound bandwidth spike might be a backup upload, CDN pull, or someone scraping your media folder. Inbound spikes might be traffic surges or DDoS probing.
Correlate with web server logs. Network alone does not tell you if the traffic is legitimate.
Patterns that map to real causes
- Daily stair-steps at the same hour: cron (backups, scans, imports)
- Spike with traffic spike: maybe you need scale, maybe you need caching
- Slow creep over weeks: database growth, log files, backup retention
- Spike after deploy: code change, new plugin, PHP version shift
- Random noise on shared hosting: neighbor activity you do not control
What to do before upgrading hardware
- Identify which metric spiked (CPU, RAM, disk I/O, network)
- Match timestamp to cron, deploys, marketing campaigns
- Check error logs and slow queries for that window
- Reproduce on staging if possible
- Fix obvious waste (debug plugins, backup to same disk, runaway cron)
Upgrading a VPS without fixing a runaway job is paying rent on the same fire. We see this often on sites that needed database cleanup more than another core.
Monitoring that helps
Panel graphs are a start. Application-level checks (homepage, checkout URL, form POST) catch "CPU looks fine but site is down" cases. We wrote about alert choice in uptime monitoring: what to actually alert on.
Keep 30 days of history if you can. "It was slow last week" needs a graph, not memory.
Bottom line
Resource graphs tell stories when you read shape, duration, and correlation with deploys and cron. Bursts are often fine. Sustained pain plus user complaints deserves investigation before you click upgrade.
Want help reading your VPS graphs on Canadian hosting? Talk with Swift Host. We will translate the spikes into fixable causes, not just bigger invoices.
