top & htop

top - 15:42:03 load average: 2.15, 1.87, 1.92 # 1min 5min 15min # Interpreting load: # Cores: 4 # Load 2.15 = 2.15 tasks queued on average # Load/Cores = 2.15/4 = 53% CPU utilization equivalent # Load consistently > cores = CPU saturated # Load rising over time = worsening problem # Load high but CPU idle = I/O bound (waiting for disk/network) # Better load check: uptime # 15:42:03 up 30 days, load average: 2.15, 1.87, 1.92

%Cpu(s): 12.5 us, 3.2 sy, 0.0 ni, 80.1 id, 4.1 wa, 0.0 hi, 0.1 si, 0.0 st # us (user) = CPU time in user-space code (your app) # sy (system) = CPU time in kernel (syscalls, interrupts) # ni (nice) = user space with positive nice value (low priority) # id (idle) = CPU idle, nothing to do # wa (iowait) = CPU idle but waiting for I/O to complete # hi (hardware interrupt) = time handling hardware IRQs # si (software interrupt) = time in softirqs (network, timers) # st (steal) = CPU stolen by hypervisor (in VMs) # Diagnosis patterns: # High us + low id: CPU bound (your code uses too much CPU) # High sy + low us: too many syscalls or kernel work # High wa: disk or network I/O bottleneck # High st: VM is over-committed on host # us + sy > 80%: system under CPU pressure

PID VIRT RES SHR S %CPU %MEM TIME+ COMMAND 1234 1.2g 345m 89m S 2.3 4.3 12:34 nginx # VIRT (Virtual memory): # Total virtual address space claimed by the process. # Includes code, data, shared libs, mmapped files, even unmapped anonymous. # Usually much larger than actual RAM used. Mostly meaningless. # RES (Resident Set Size): # Physical RAM actually in use RIGHT NOW. # This is what you care about for "how much RAM does this process use" # Includes shared libraries (counted even if shared with other processes) # SHR (Shared memory): # Portion of RES that's shared with other processes. # Shared libraries, shared mmap files. # If 50 processes share libc, SHR shows libc's size in each. # True "private" RAM = RES - SHR # Total system RAM for process = RES (but SHR counted once for system total) # Better memory view: cat /proc/1234/smaps_rollup # Rss: 354432 kB (same as RES) # Pss: 134567 kB (proportional share — SHR divided among users) # Private_Clean: 45678 kB # Private_Dirty: 123456 kB ← this process exclusively owns this

# Install: apt install htop # htop advantages over top: # - Per-CPU bar graphs (see which CPUs are busy) # - Better color coding (user/system/IO wait each different color) # - Easy sorting: click column header or F6 # - Tree view: F5 (shows parent-child relationships) # - Kill without leaving: select process, F9 # - Search: F3 # - Scroll horizontally: see full command # Key htop columns: # PID = process ID # USER = owner # PRI/NI = priority/nice (-20 high priority, 19 low) # VIRT/RES/SHR = same as top # S = state (R=running, S=sleep, D=uninterruptible, Z=zombie) # CPU% = CPU usage (can exceed 100% on multi-core) # MEM% = % of total RAM # TIME+ = total CPU time used # Interactive keys: # u = filter by user # k = kill process # r = renice (change priority) # s = strace the process # l = lsof the process (list open files) # I = invert sort order