feat: Introduce measure_bandwidth.sh script, install perfquery, and add the script to the Docker image for RDMA bandwidth monitoring.

Dockerfile

@@ -127,6 +127,7 @@ COPY scripts/99-toolbox-banner.sh /etc/profile.d/99-toolbox-banner.sh
 COPY scripts/zz-venv-last.sh /etc/profile.d/zz-venv-last.sh
 COPY scripts/start_vllm.py /opt/start-vllm
 COPY scripts/start_vllm_cluster.py /opt/start-vllm-cluster
+COPY scripts/measure_bandwidth.sh /opt/measure_bandwidth.sh
 COPY scripts/cluster_manager.py /opt/cluster_manager.py
 COPY scripts/models.py /opt/models.py
 

scripts/install_deps.sh

@@ -8,5 +8,5 @@ dnf -y install --setopt=install_weak_deps=False --nodocs \
   gcc gcc-c++ binutils make ffmpeg-free \
   cmake ninja-build aria2c tar xz vim nano dialog \
   libdrm-devel zlib-devel openssl-devel pgrep \
-  numactl-devel gperftools-libs iproute libibverbs-utils patch perftest ping iperf3 \
+  numactl-devel gperftools-libs iproute libibverbs-utils patch perftest ping iperf3 perfquery \
   && dnf clean all && rm -rf /var/cache/dnf/*

scripts/measure_bandwidth.sh

@@ -0,0 +1,18 @@
+#!/usr/bin/env bash
+
+while true; do
+  A_IN=$(rdma statistic | awk '/ip4InOctets/ {print $2}')
+  A_OUT=$(rdma statistic | awk '/ip4OutOctets/ {print $2}')
+  sleep 1
+  B_IN=$(rdma statistic | awk '/ip4InOctets/ {print $2}')
+  B_OUT=$(rdma statistic | awk '/ip4OutOctets/ {print $2}')
+
+  RX=$(( (B_IN - A_IN) * 8 ))
+  TX=$(( (B_OUT - A_OUT) * 8 ))
+
+  printf "%s RDMA RX: %7sbit/s  TX: %7sbit/s  SUM: %7sbit/s\n" \
+    "$(date +%T)" \
+    "$(numfmt --to=iec $RX)" \
+    "$(numfmt --to=iec $TX)" \
+    "$(numfmt --to=iec $((RX+TX)))"
+done