Heathrow Closed: Flight Suspension Expected to Continue for the Coming Days, Says London Airport Manager
British Airways estimated that 85% of its scheduled flights would operate on Saturday, but all flights were delayed. By 7:00 a.m. GMT, most departures had proceeded as expected, but of the arrivals, nine of the first 20 flights scheduled to land were canceled.
Somewhere between versions 0.3.31-2 and 0.3.31-7, a change occurred that forcibly changes the listening interface from external to 127.0.0.1 and disables CORS.
Since LM Studio supports hot-reload of the HTTP server configuration (at least when changing cors and networkInterface), you can make the change without restarting (it will work once):
%h/.lmstudio/.internal/http-server-config.json
jq '.cors = true | .networkInterface = "0.0.0.0"' ~/.lmstudio/.internal/http-server-config.json | sponge ~/.lmstudio/.internal/http-server-config.json
remember to install sponge, which is included in the moreutils package.
To make the change automatic, modify the startup script:
$HOME/.config/systemd/user/lm-studio.service
[Unit]
Description=LM Studio Service
After=network.target
[Service]
Type=simple
ExecStart=/usr/bin/xvfb-run -a --server-args="-screen 0 1920x1080x24" %h/llm/lmstudio --run-as-service
# 1. Start the HTTP server
ExecStartPost=/bin/bash -c 'sleep 15 && exec lms server start'
# 2. Apply required settings (after the server has already started)
ExecStartPost=/bin/bash -c ' \
sleep 2 && \
jq ".cors = true | .networkInterface = \"0.0.0.0\"" \
"%h/.lmstudio/.internal/http-server-config.json" \
> "%h/.lmstudio/.internal/http-server-config.json.tmp" && \
mv "%h/.lmstudio/.internal/http-server-config.json.tmp" \
"%h/.lmstudio/.internal/http-server-config.json" \
'
Restart=always
RestartSec=10
Environment=PATH=%h/.local/bin:/usr/local/bin:/usr/bin:/bin:%h/.lmstudio/bin
Environment=DISPLAY=:99
WorkingDirectory=%h/llm
[Install]
WantedBy=default.target
Apply the changes and restart the service (without sudo!):
systemctl --user daemon-reload
systemctl --user stop lm-studio.service
systemctl --user start lm-studio.service
Achten Sie auf die Optimierung der Betriebssystemparameter für optimale Leistung.
Hier ein Beispiel aus meinem System (ein Laptop mit einer über Oculink angeschlossenen Grafikkarte):
free -h
total used free shared buff/cache available
Mem: 37Gi 4.9Gi 29Gi 240Mi 3.5Gi 32Gi
Swap: 8.0Gi 0B 8.0Gi
cat /proc/meminfo | grep -E 'MemTotal|MemAvailable'
MemTotal: 39223064 kB
MemAvailable: 34073484 kB
VRAM wird separat angezeigt – sie ist nicht Teil von MemTotal.
nvidia-smi
Fri Nov 21 12:24:47 2025
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 580.82.09 Driver Version: 580.82.09 CUDA Version: 13.0 |
+-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 NVIDIA GeForce RTX 5060 Ti Off | 00000000:01:00.0 Off | N/A |
| 0% 44C P8 8W / 180W | 13MiB / 16311MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
| 0 N/A N/A 5314 G /usr/bin/gnome-shell 2MiB |
+-----------------------------------------------------------------------------------------+
Überprüfen Sie die Einstellungen für overcommit und swappiness:
cat /proc/sys/vm/overcommit_memory
cat /proc/sys/vm/overcommit_ratio
cat /proc/sys/vm/swappiness
Empfohlene Werte:
2 # Limit = (RAM * ratio/100) + Swap
80 # Dies ist der Ratio-Wert
10 # Nicht höher als 20 – das ist der Anteil der verbleibenden RAM, bei dem Swap aktiviert wird; der Swap kann 8 GiB betragen, wenn die Betriebssystem-Hibernation nicht genutzt wird, dann entspricht der Swap der Größe der RAM
Ausführbare Befehle:
sudo sysctl vm.overcommit_ratio=80
echo 'vm.overcommit_ratio=80' | sudo tee -a /etc/sysctl.d/99-ml-workstation.conf
sudo sysctl vm.swappiness=10
echo 'vm.swappiness=10' | sudo tee -a /etc/sysctl.d/99-ml-workstation.conf
WICHTIG: Bei LLMs (insbesondere llama.cpp) kann THP = always zu Pausen (Verzögerungen von einer Minute nach dem Systemstart) führen.
cat /sys/kernel/mm/transparent_hugepage/enabled
# Sollte sein: [always] madvise never
Beim Arbeiten mit Speicher (z. B. wenn LLM-Modelle auf SSD geschrieben werden), ist die writeback-Einstellung wichtig:
cat /proc/sys/vm/dirty_ratio
cat /proc/sys/vm/dirty_background_ratio
cat /proc/sys/vm/dirty_expire_centisecs
Optimale Werte für NVMe:
10
5
1000 (das sind 10 Sekunden)
echo 'vm.dirty_ratio=10' | sudo tee -a /etc/sysctl.d/99-ml-workstation.conf
echo 'dirty_background_ratio=5' | sudo tee -a /etc/sysctl.d/99-ml-workstation.conf
echo 'vm.dirty_expire_centisecs=1000' | sudo tee -a /etc/sysctl.d/99-ml-workstation.conf
Zwischen den Starts von lmstudio wird empfohlen, den Speicher-Cache zu leeren:
sync && echo 2 | sudo tee /proc/sys/vm/drop_caches