server_strain

A portable system strain indicator for Linux. Run it, get one integer back indicating the realtive strain on the server.

0 = near idle  |  100 = fully utilized, no bottlenecks  |  > 100 = over utilization

$ ./server_strain.sh
42

---

Scale

Range	Meaning
0 to 30	Near idle
30 to 70	Moderate load
70 to 100	Heavy but healthy utilization, within limits
> 100	Bottleneck detected, something is saturated

---

Differences from system Load Average

Load average is a single raw number the kernel maintains, the exponentially weighted average of runnable and uninterruptible tasks over 1, 5, and 15 minute windows. Its problems:

1. It has no upper bound and no context. A load of 8.0 is fine on a 32-core machine but very bad on a 2-core one.

2. It only counts tasks waiting for CPU or disk. A system melting under memory pressure or saturated network I/O can show a deceptively low load average.

3. It tells you how many things are waiting, not how close to the edge you are. server_strain is a normalized, multi-signal score designed to answer a different question: "what fraction of this system's total capacity is being consumed right now?"

A practical example: a 40-core server with load average 4.0 looks fine by load average (only 10% of cores busy). But if that same machine has 85% memory used and 25% iowait, server_strain would be around 90+. This is much closer to the truth that the system is under real pressure. Load average would give you no hint of that.

Usage

# Default: 1-second CPU sample window
./server_strain.sh

# Custom sample interval (seconds) more stable on bursty workloads
./server_strain.sh 5

No root required. The script only reads from /proc and calls awk / nproc, which are present on essentially every Linux distribution.

---

How it works

server_strain collects four signals, blends them into a base score, then adds penalty points for any active bottlenecks.

Step 1: CPU utilization and I/O wait

The kernel exposes cumulative CPU jiffie counts in /proc/stat. The script reads that file twice, separated by the sample interval, and computes the delta:

cpu_busy%  =  100 × (total_delta − idle_delta − iowait_delta) / total_delta
iowait%    =  100 × iowait_delta / total_delta

Separating iowait matters: a CPU blocked on disk shows up as "idle" in the busy figure but is captured here as iowait, so no signal is lost.

Step 2: Memory pressure

mem_used%  =  100 × (MemTotal − MemAvailable) / MemTotal

MemAvailable not MemFree is used because the kernel reclaims page cache on demand. This avoids false pressure warnings on systems that aggressively cache disk reads.

Swap usage is tracked separately for the bottleneck penalty in step 5.

Step 3: Normalized load average

load%  =  (load_1m / cores) × 100

The 1-minute load average from /proc/loadavg is divided by the logical CPU core count. A value of 100% means exactly one runnable task per core, fully committed but not over-scheduled.

Step 4: Base strain (weighted average)

The four signals are blended into a 0 to 100 base score:

Signal	Weight	Rationale
CPU busy %	40%	Primary throughput indicator
Normalized load %	30%	Scheduling demand, including disk/network wait states
Memory used %	20%	Working-set pressure
I/O wait %	10%	Disk latency (partially reflected in load already)

base_strain  =  cpu×0.40  +  load×0.30  +  mem×0.20  +  iowait×0.10

On a perfectly idle system this approaches 0. On a system where every resource is fully saturated with nothing queuing, it approaches 100.

Step 5: Bottleneck penalties

When resources are over-committed, penalty points are added on top of the base score, pushing server_strain above 100:

Condition	Penalty formula	What it signals
Load average > core count	+(excess/cores) × 30	Tasks queuing for CPU, scheduler backlogged
I/O wait > 20%	+(iowait − 20) / 2	Disk is a throughput bottleneck
Swap used > 5%	+(swap_pct − 5) / 2	Physical RAM exhausted, actively paging

A strain of 120, for example, means the system is not just fully utilized but is actively building up queues and response times will be degrading.

---

Dependencies

Dependency	Purpose
bash	Shell interpreter
awk	Field parsing and floating-point arithmetic
nproc	Core count (falls back to /proc/cpuinfo)
/proc/stat	CPU jiffie counters
/proc/meminfo	Memory and swap figures
/proc/loadavg	Load average

All dependencies are part of the Linux base system. No packages to install. Should work on any modern distro.