FeedSim V2: update job config, install and run script (facebookresearch#364)

Summary:

- Install DLRM deps and download the model in install scripts
- Create a new `feedsim_dlrm` benchmark job for this new workload

Reviewed By: YifanYuan3

Differential Revision: D89341996

Author: excelle08

benchpress/config/jobs.yml

@@ -538,9 +538,51 @@
       options:
         is_move: true
         after:
+          - '/tmp/feedsim_log.txt'
           - 'benchmarks/feedsim/feedsim_results.txt'
           - 'benchmarks/feedsim/src/perf.data'
 
+- name: feedsim_dlrm
+  benchmark: feedsim
+  description: >
+    Aggregator like workload with DLRM (Deep Learning Recommendation Model)
+    inference. Uses async I/O for improved throughput on high-core CPUs.
+    Finds maximum QPS that system can sustain while keeping
+    95th percentile latency <= 500 msecs.
+  args:
+    - '-p {port}'
+    - '-o {output}'
+    - '--async-io'
+    - '--io-dist={io_dist}'
+    - '--io-mean={io_mean}'
+    - '--workload={workload}'
+    - '--dlrm-model={dlrm_model}'
+    - '--dlrm-batch-size={dlrm_batch_size}'
+    - '--dlrm-threads={dlrm_threads}'
+  vars:
+    - 'port=11222'
+    - 'output=feedsim_results.txt'
+    - 'io_dist=fixed'
+    - 'io_mean=200'
+    - 'workload=dlrm'
+    - 'dlrm_model=models/dlrm_small.pt'
+    - 'dlrm_batch_size=256'
+    - 'dlrm_threads=8'
+  hooks:
+    - hook: cpu-mpstat
+      options:
+        args:
+          - '-u'   # utilization
+          - '1'    # second interval
+    - hook: copymove
+      options:
+        is_move: true
+        after:
+          - '/tmp/feedsim_log.txt'
+          - 'benchmarks/feedsim/feedsim_results.txt'
+          - 'benchmarks/feedsim/src/perf.data'
+          - '/tmp/feedsim_log.txt'
+
 - name: feedsim_autoscale
   benchmark: feedsim_autoscale
   description: >
@@ -568,6 +610,7 @@
           - 'benchmarks/feedsim/feedsim_results*.txt'
           - 'benchmarks/feedsim/feedsim-multi-inst-*.log'
           - 'benchmarks/feedsim/src/perf.data'
+          - '/tmp/feedsim_log.txt'
 
 - name: feedsim_autoscale_mini
   benchmark: feedsim_autoscale
@@ -613,6 +656,7 @@
           - 'benchmarks/feedsim/feedsim-multi-inst-*.log'
           - 'benchmarks/feedsim/src/perf.data'
           - 'benchmarks/feedsim/breakdown.csv'
+          - '/tmp/feedsim_log.txt'
 
 - name: feedsim_autoscale_arm
   benchmark: feedsim_autoscale
@@ -643,6 +687,7 @@
           - 'benchmarks/feedsim/feedsim_results*.txt'
           - 'benchmarks/feedsim/feedsim-multi-inst-*.log'
           - 'benchmarks/feedsim/src/perf.data'
+          - '/tmp/feedsim_log.txt'
 
 - name: feedsim_autoscale_arm_mini
   benchmark: feedsim_autoscale
@@ -685,6 +730,7 @@
           - 'benchmarks/feedsim/feedsim-multi-inst-*.log'
           - 'benchmarks/feedsim/src/perf.data'
           - 'benchmarks/feedsim/breakdown.csv'
+          - '/tmp/feedsim_log.txt'
 
 
 - benchmark: spark_standalone

packages/feedsim/alternative_install_feedsim_aws.sh

@@ -12,6 +12,8 @@ FEEDSIM_ROOT=$(cd "$(dirname "${BASH_SOURCE[0]}")" &>/dev/null && pwd -P)
 BENCHPRESS_ROOT="$(readlink -f "$FEEDSIM_ROOT/../..")"
 FEEDSIM_ROOT_SRC="${BENCHPRESS_ROOT}/benchmarks/feedsim"
 FEEDSIM_THIRD_PARTY_SRC="${FEEDSIM_ROOT_SRC}/third_party"
+LIBTORCH_VERSION="2.1.0"
+DLRM_MODEL_URL="https://github.com/facebookresearch/DCPerf-datasets/releases/download/feedsim-dlrm/dlrm_small.tar.gz"
 echo "BENCHPRESS_ROOT is ${BENCHPRESS_ROOT}"
 
 source "${BENCHPRESS_ROOT}/packages/common/os-distro.sh"
@@ -48,7 +50,8 @@ fi
 dnf install -y bc ninja-build flex bison git texinfo binutils-devel \
     libsodium-devel libunwind-devel bzip2-devel double-conversion-devel \
     libzstd-devel lz4-devel xz-devel snappy-devel libtool bzip2 openssl-devel \
-    zlib-devel libdwarf libdwarf-devel libaio-devel libatomic patch jq xxhash xxhash-devel
+    zlib-devel libdwarf libdwarf-devel libaio-devel libatomic patch jq \
+    xxhash xxhash-devel unzip
 
 # Creates feedsim directory under benchmarks/
 mkdir -p "${BENCHPRESS_ROOT}/benchmarks/feedsim"
@@ -176,6 +179,48 @@ fi
 
 msg "Installing third-party dependencies ... DONE"
 
+# Installing LibTorch for DLRM support
+msg "Installing LibTorch for DLRM support..."
+cd "${FEEDSIM_THIRD_PARTY_SRC}"
+
+ARCH="$(uname -m)"
+if [ "$ARCH" = "x86_64" ]; then
+    LIBTORCH_URL="https://download.pytorch.org/libtorch/cpu/libtorch-cxx11-abi-shared-with-deps-${LIBTORCH_VERSION}%2Bcpu.zip"
+elif [ "$ARCH" = "aarch64" ]; then
+    msg "WARNING: Pre-built LibTorch for ARM64 may not be available."
+    msg "Attempting to download CPU version..."
+    LIBTORCH_URL="https://download.pytorch.org/libtorch/cpu/libtorch-cxx11-abi-shared-with-deps-${LIBTORCH_VERSION}%2Bcpu.zip"
+else
+    die "Unsupported architecture: ${ARCH}"
+fi
+
+if ! [ -d "libtorch" ]; then
+    msg "Downloading LibTorch ${LIBTORCH_VERSION}..."
+    wget "${LIBTORCH_URL}" -O libtorch.zip
+    msg "Extracting LibTorch..."
+    unzip -q libtorch.zip
+    rm libtorch.zip
+    msg "LibTorch installed to ${FEEDSIM_THIRD_PARTY_SRC}/libtorch"
+else
+    msg "[SKIPPED] LibTorch already installed"
+fi
+
+# Download DLRM model
+msg "Downloading DLRM model..."
+DLRM_MODEL_DIR="${FEEDSIM_ROOT_SRC}/models"
+mkdir -p "${DLRM_MODEL_DIR}"
+
+if ! [ -f "${DLRM_MODEL_DIR}/dlrm_small.pt" ]; then
+    msg "Downloading DLRM model from ${DLRM_MODEL_URL}..."
+    wget "${DLRM_MODEL_URL}" -O "${DLRM_MODEL_DIR}/dlrm_small.tar.gz"
+    msg "Extracting DLRM model..."
+    tar -xzf "${DLRM_MODEL_DIR}/dlrm_small.tar.gz" -C "${DLRM_MODEL_DIR}"
+    rm "${DLRM_MODEL_DIR}/dlrm_small.tar.gz"
+    msg "DLRM model installed to ${DLRM_MODEL_DIR}"
+else
+    msg "[SKIPPED] DLRM model already installed"
+fi
+
 
 # Installing FeedSim
 cd "${FEEDSIM_ROOT_SRC}"
@@ -203,18 +248,32 @@ fi
 
 mkdir -p build && cd build/
 
-# Build FeedSim
+# Build FeedSim with DLRM support
 FS_CFLAGS="${BP_CFLAGS:--O3 -DNDEBUG}"
 FS_CXXFLAGS="${BP_CXXFLAGS:--O3 -DNDEBUG }"
 FS_LDFLAGS="${BP_LDFLAGS:-} -latomic -Wl,--export-dynamic"
 
 cmake -G Ninja \
     -DCMAKE_BUILD_TYPE=Release \
-    -DCMAKE_C_COMPILER="$BP_CC" \
-    -DCMAKE_CXX_COMPILER="$BP_CXX" \
+    -DCMAKE_C_COMPILER="${BP_CC:-gcc}" \
+    -DCMAKE_CXX_COMPILER="${BP_CXX:-g++}" \
     -DCMAKE_C_FLAGS_RELEASE="$FS_CFLAGS" \
     -DCMAKE_CXX_FLAGS_RELEASE="$FS_CXXFLAGS -DFMT_HEADER_ONLY=1" \
     -DCMAKE_EXE_LINKER_FLAGS_RELEASE="$FS_LDFLAGS" \
+    -DFEEDSIM_USE_DLRM=ON \
+    -DTorch_DIR="${FEEDSIM_THIRD_PARTY_SRC}/libtorch/share/cmake/Torch" \
+    -DCMAKE_PREFIX_PATH="${FEEDSIM_THIRD_PARTY_SRC}/libtorch" \
     ../
 
 ninja -v -j1
+
+msg ""
+msg "=== FeedSim Installation Complete ==="
+msg ""
+msg "To run FeedSim with DLRM workload:"
+msg "  cd ${FEEDSIM_ROOT_SRC}"
+msg "  ./run.sh --workload=dlrm --dlrm-model=${DLRM_MODEL_DIR}/dlrm_small.pt"
+msg ""
+msg "Or use the standard PageRank workload:"
+msg "  ./run.sh"
+msg ""