Add Multi-Architecture Support (x86_64, arm64) for Benchmarks (#227)

* [Feat] Added architecture parameter in config file [x86_64 or arm64]

* [Fix] Change python version to 3.7

* [Feat] Changed python installter to check for architecture and install the package accordingly.

* [Feat] Get architecture and pass that as paramter while creating lambda function.

* [Feat] Pass architecture paramter while running the docker container.

* [Feat] Add architecture in config deserializer.

* [Feat] Add architecture support.

* [Feat] Add requirement for arm for 210.thumbnailer

* [Feat] Add requirement for arm for 501.graph-pagerank

* [Feat] Add requirement for arm for 502.graph-mst

* [Feat] Add requirement for arm for 503.graph-bfs

* [Feat] Add requirement for arm for 504.dna-visualisation

* [Feat] Add architecture support build for ffmpeg for 220.vide-processing

* Fail early in case of architecutre is not suppoerted in the platform.

* Add Architecture paramter when installing npm packages

* Testing NodeJS ( Multi architecture support)

* Upgrage version of sharp as previous 0.28 version didnt had the  prebuilt arm libraries

* Add approprivate version for python3.9 arm

* [dev] Linting

* [Feat] Fix order of parameters

* [Feat] Support AWS arm64 architecture also when updating functions

* [Feat] Add architecture to function name on AWS

* [Feat] Separate cached code packages and output directories based on architecture

* [Feat] Rename x86_64 to a simpler x64

* [Feat] Fix incorrect cache creation

* [system] Fix major bug in cache handling - we failed to update size of new code package

* [Feat] Adapt regression system to work with regression

* [Feat] Add architecture to code packages uploaded to cloud storage

* [dev] Linting

* [Feat] Fix paths to code packages in S3

* [Feat] Support Python 3.10 on ARM

* [Feat] Fix build of 504.dna-visualization on ARM where one package blows up code package size

* [aws] Implement backoff for AWS API that has a low limit on the number of concurrent requests

* [Feat] Expand docs

---------

Co-authored-by: prajinkhadka <prawjeenkhadka@gmail.com>

benchmarks/200.multimedia/210.thumbnailer/nodejs/package.json.16

@@ -5,6 +5,6 @@
   "author": "",
   "license": "",
   "dependencies": {
-    "sharp": "^0.28"
+    "sharp": "^0.32"
   }
 }

benchmarks/200.multimedia/210.thumbnailer/python/requirements.txt.arm.3.8

@@ -0,0 +1 @@
+Pillow==10.0.0

benchmarks/200.multimedia/210.thumbnailer/python/requirements.txt.arm.3.9

@@ -0,0 +1 @@
+Pillow==10.0.0

benchmarks/200.multimedia/220.video-processing/init.sh

@@ -2,13 +2,23 @@
 
 DIR=$1
 VERBOSE=$2
+TARGET_ARCHITECTURE=$3
+
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" >/dev/null 2>&1 && pwd)"
-wget -q https://johnvansickle.com/ffmpeg/releases/ffmpeg-release-amd64-static.tar.xz -P ${DIR}
+
+if [[ "${TARGET_ARCHITECTURE}" == "arm64" ]]; then
+    FFMPEG_URL="https://johnvansickle.com/ffmpeg/releases/ffmpeg-release-arm64-static.tar.xz"
+else
+    FFMPEG_URL="https://johnvansickle.com/ffmpeg/releases/ffmpeg-release-amd64-static.tar.xz"
+fi
+
+wget -q ${FFMPEG_URL} -P ${DIR}
+
 pushd ${DIR} >/dev/null
-tar -xf ffmpeg-release-amd64-static.tar.xz
+tar -xf ffmpeg-release-*-static.tar.xz
 rm *.tar.xz
 mv ffmpeg-* ffmpeg
-rm ffmpeg/ffprobe
+rm -f ffmpeg/ffprobe 
 # make the binary executable
 chmod 755 ffmpeg/ffmpeg
 popd >/dev/null

benchmarks/500.scientific/501.graph-pagerank/python/requirements.txt.arm.3.8

@@ -0,0 +1 @@
+python-igraph==0.11.4

benchmarks/500.scientific/501.graph-pagerank/python/requirements.txt.arm.3.9

@@ -0,0 +1 @@
+python-igraph==0.11.4

benchmarks/500.scientific/502.graph-mst/python/requirements.txt.arm.3.8

@@ -0,0 +1 @@
+python-igraph==0.11.4

benchmarks/500.scientific/502.graph-mst/python/requirements.txt.arm.3.9

@@ -0,0 +1 @@
+python-igraph==0.11.4

benchmarks/500.scientific/503.graph-bfs/python/requirements.txt.arm.3.8

@@ -0,0 +1 @@
+python-igraph==0.11.4

benchmarks/500.scientific/503.graph-bfs/python/requirements.txt.arm.3.9

@@ -0,0 +1 @@
+python-igraph==0.11.4

benchmarks/500.scientific/504.dna-visualisation/python/package.sh

@@ -0,0 +1,21 @@
+# Some dependencies have broken wheels
+# For example, pandas wheel for 3.10 and arm ships libraries
+# for all Linux versions and all Python versions
+# This is too much for Lambda
+
+PACKAGE_DIR=$1
+echo "Original size $(du -sh $1 | cut -f1)"
+
+CUR_DIR=$(pwd)
+cd $1
+
+# remove libraries for musl
+find . -name "*aarch64-linux-musl.so" | xargs rm
+
+version=$(echo "${PYTHON_VERSION}" | sed 's/\.//g')
+echo "versions ${PYTHON_VERSION} ${version}"
+# remove libraries for other Python versions
+find . -name "*aarch64-linux-gnu.so" | grep -v ${version} | xargs rm
+
+cd ${CUR_DIR}
+echo "Stripped size $(du -sh $1 | cut -f1)"

benchmarks/500.scientific/504.dna-visualisation/python/requirements.txt.arm.3.8

@@ -0,0 +1 @@
+squiggle==0.3.1

benchmarks/500.scientific/504.dna-visualisation/python/requirements.txt.arm.3.9

@@ -0,0 +1 @@
+squiggle==0.3.1

config/example.json

@@ -3,10 +3,11 @@
     "deployment": "openwhisk",
     "update_code": false,
     "update_storage": false,
-    "download_results": false,
+    "download_results": false, 
+    "architecture": "arm64", 
     "runtime": {
-      "language": "python",
-      "version": "3.7"
+      "language": "nodejs",
+      "version": "16"
     },
     "type": "invocation-overhead",
     "perf-cost": {

config/systems.json

@@ -54,7 +54,8 @@
           "packages": []
         }
       }
-    }
+    },
+    "architecture": ["x64"]
   },
   "aws": {
     "languages": {
@@ -93,7 +94,8 @@
           }
         }
       }
-    }
+    },
+    "architecture": ["x64", "arm64"]
   },
   "azure": {
     "languages": {
@@ -145,7 +147,8 @@
       "manage": {
         "username": "docker_user"
       }
-    }
+    },
+    "architecture": ["x64"]
   },
   "gcp": {
     "languages": {
@@ -196,7 +199,8 @@
           }
         }
       }
-    }
+    },
+    "architecture": ["x64"]
   },
   "openwhisk": {
     "languages": {
@@ -244,6 +248,7 @@
           }
         }
       }
-    }
+    },
+    "architecture": ["x64"]
   }
 }

dockerfiles/nodejs_installer.sh

@@ -1,6 +1,18 @@
 #!/bin/bash
 
-if [ -f $FILE ]; then
+if [ -f /nvm/nvm.sh ]; then
   . /nvm/nvm.sh
 fi
-cd /mnt/function && npm install && rm -rf package-lock.json
+#
+cd /mnt/function
+
+if [ "${TARGET_ARCHITECTURE}" == "arm64" ]; then
+        npm install --arch=arm64
+elif [ "${TARGET_ARCHITECTURE}" = "x64" ]; then
+        npm install --arch=x64
+else
+        echo "Unsupported architecture: $TARGET_ARCHITECTURE"
+        exit 1
+fi
+
+rm -rf package-lock.json

dockerfiles/python_installer.sh

@@ -1,5 +1,28 @@
 #!/bin/bash
 
-cd /mnt/function\
-  && if test -f "requirements.txt.${PYTHON_VERSION}"; then pip3 -q install -r requirements.txt -r requirements.txt.${PYTHON_VERSION} -t .python_packages/lib/site-packages ; else pip3 -q install -r requirements.txt -t .python_packages/lib/site-packages ; fi\
-  && if test -f "${SCRIPT_FILE}"; then /bin/bash ${SCRIPT_FILE} .python_packages/lib/site-packages ; fi
+cd /mnt/function
+
+PLATFORM_ARG=""
+if [[ "${TARGET_ARCHITECTURE}" == "arm64" ]]; then
+  PLATFORM_ARG="--platform manylinux_2_17_aarch64 --only-binary=:all:"
+fi
+
+if [[ "${TARGET_ARCHITECTURE}" == "arm64" ]] && [[ -f "requirements.txt.arm.${PYTHON_VERSION}" ]]; then
+
+  pip3 -q install ${PLATFORM_ARG} -r requirements.txt.arm.${PYTHON_VERSION} -t .python_packages/lib/site-packages
+
+elif [[ -f "requirements.txt.${PYTHON_VERSION}" ]]; then
+
+  pip3 -q install ${PLATFORM_ARG} -r requirements.txt.${PYTHON_VERSION} -t .python_packages/lib/site-packages
+
+else
+
+  pip3 -q install ${PLATFORM_ARG} -r requirements.txt -t .python_packages/lib/site-packages
+
+fi
+
+if [[ -f "${SCRIPT_FILE}" ]]; then
+  /bin/bash ${SCRIPT_FILE} .python_packages/lib/site-packages
+fi
+
+

docs/benchmarks.md

@@ -1,24 +1,27 @@
 
 ## Benchmark Applications
 
-| Type 		   | Benchmark           | Languages          | Description          |
-| :---         | :---:               | :---:              | :---:                |
-| Webapps      | 110.dynamic-html    | Python, Node.js    | Generate dynamic HTML from a template. |
-| Webapps      | 120.uploader    | Python, Node.js    | Uploader file from provided URL to cloud storage. |
-| Multimedia      | 210.thumbnailer    | Python, Node.js    | Generate a thumbnail of an image. |
-| Multimedia      | 220.video-processing    | Python    | Add a watermark and generate gif of a video file. |
-| Utilities      | 311.compression    | Python   | Create a .zip file for a group of files in storage and return to user to download. |
-| Inference      | 411.image-recognition    | Python    | Image recognition with ResNet and pytorch. |
-| Scientific      | 501.graph-pagerank    | Python    | PageRank implementation with igraph. |
-| Scientific      | 502.graph-mst    | Python    | Minimum spanning tree (MST)  implementation with igraph. |
-| Scientific      | 503.graph-bfs    | Python    | Breadth-first search (BFS) implementation with igraph. |
-| Scientific      | 504.dna-visualisation    | Python   | Creates a visualization data for DNA sequence. |
+| Type 		   | Benchmark           | Languages          | Architecture       |  Description |
+| :---         | :---:               | :---:              | :---:                | :---:                |
+| Webapps      | 110.dynamic-html    | Python, Node.js    | x64, arm64 | Generate dynamic HTML from a template. |
+| Webapps      | 120.uploader    | Python, Node.js    | x64, arm64 | Uploader file from provided URL to cloud storage. |
+| Multimedia      | 210.thumbnailer    | Python, Node.js    | x64, arm64 | Generate a thumbnail of an image. |
+| Multimedia      | 220.video-processing    | Python    | x64, arm64 | Add a watermark and generate gif of a video file. |
+| Utilities      | 311.compression    | Python   | x64, arm64 | Create a .zip file for a group of files in storage and return to user to download. |
+| Inference      | 411.image-recognition    | Python    | x64 | Image recognition with ResNet and pytorch. |
+| Scientific      | 501.graph-pagerank    | Python    | x64, arm64 | PageRank implementation with igraph. |
+| Scientific      | 502.graph-mst    | Python    | x64, arm64 | Minimum spanning tree (MST)  implementation with igraph. |
+| Scientific      | 503.graph-bfs    | Python    | x64, arm64 | Breadth-first search (BFS) implementation with igraph. |
+| Scientific      | 504.dna-visualisation    | Python   | x64, arm64 | Creates a visualization data for DNA sequence. |
 
 Below, we discuss the most important implementation details of each benchmark. For more details on benchmark selection and their characterization, please refer to [our paper](../README.md#publication).
 
 > [!NOTE] 
 > Benchmarks whose number starts with the digit 0, such as `020.server-reply` are internal microbenchmarks used by specific experiments. They are not intended to be directly invoked by users.
 
+> [!NOTE] 
+> ARM architecture is supported on AWS Lambda only.
+
 > [!WARNING] 
 > Benchmark 411.image-recognition contains PyTorch which is often too large to fit into a code package. Up to Python 3.7, we can directly ship the dependencies. For Python 3.8, we use an additional zipping step that requires additional setup during the first run, making cold invocations slower. Warm invocations are not affected.
 

sebs.py

@@ -91,12 +91,19 @@ def common_params(func):
         type=click.Choice(["azure", "aws", "gcp", "local", "openwhisk"]),
         help="Cloud deployment to use.",
     )
+    @click.option(
+        "--architecture",
+        default=None,
+        type=click.Choice(["x64", "arm64"]),
+        help="Target architecture",
+    )
     @click.option(
         "--resource-prefix",
         default=None,
         type=str,
         help="Resource prefix to look for.",
     )
+
     @simplified_common_params
     @functools.wraps(func)
     def wrapper(*args, **kwargs):
@@ -117,6 +124,7 @@ def parse_common_params(
     deployment,
     language,
     language_version,
+    architecture,
     resource_prefix: Optional[str] = None,
     initialize_deployment: bool = True,
     ignore_cache: bool = False,
@@ -139,6 +147,7 @@ def parse_common_params(
     update_nested_dict(config_obj, ["deployment", "name"], deployment)
     update_nested_dict(config_obj, ["experiments", "update_code"], update_code)
     update_nested_dict(config_obj, ["experiments", "update_storage"], update_storage)
+    update_nested_dict(config_obj, ["experiments", "architecture"], architecture)
 
     # set the path the configuration was loaded from
     update_nested_dict(config_obj, ["deployment", "local", "path"], config)
@@ -149,7 +158,7 @@ def parse_common_params(
 
     if initialize_deployment:
         deployment_client = sebs_client.get_deployment(
-            config_obj["deployment"], logging_filename=logging_filename
+            config_obj, logging_filename=logging_filename
         )
         deployment_client.initialize(resource_prefix=resource_prefix)
     else:
@@ -342,7 +351,7 @@ def regression(benchmark_input_size, benchmark_name, **kwargs):
         sebs_client,
         config["experiments"],
         set((config["deployment"]["name"],)),
-        config["deployment"],
+        config,
         benchmark_name,
     )
 

sebs/aws/aws.py

@@ -159,6 +159,12 @@ def package_code(
 
         return os.path.join(directory, "{}.zip".format(benchmark)), bytes_size
 
+    def _map_architecture(self, architecture: str) -> str:
+
+        if architecture == "x64":
+            return "x86_64"
+        return architecture
+
     def _map_language_runtime(self, language: str, runtime: str):
 
         # AWS uses different naming scheme for Node.js versions
@@ -180,7 +186,7 @@ def create_function(self, code_package: Benchmark, func_name: str) -> "LambdaFun
         func_name = AWS.format_function_name(func_name)
         storage_client = self.get_storage()
         function_cfg = FunctionConfig.from_benchmark(code_package)
-
+        architecture = function_cfg.architecture.value
         # we can either check for exception or use list_functions
         # there's no API for test
         try:
@@ -217,7 +223,7 @@ def create_function(self, code_package: Benchmark, func_name: str) -> "LambdaFun
                 code_package_name = cast(str, os.path.basename(package))
 
                 code_bucket = storage_client.get_bucket(Resources.StorageBucketType.DEPLOYMENT)
-                code_prefix = os.path.join(benchmark, code_package_name)
+                code_prefix = os.path.join(benchmark, architecture, code_package_name)
                 storage_client.upload(code_bucket, package, code_prefix)
 
                 self.logging.info("Uploading function {} code to {}".format(func_name, code_bucket))
@@ -232,6 +238,7 @@ def create_function(self, code_package: Benchmark, func_name: str) -> "LambdaFun
                 MemorySize=memory,
                 Timeout=timeout,
                 Code=code_config,
+                Architectures=[self._map_architecture(architecture)],
             )
 
             lambda_function = LambdaFunction(
@@ -283,25 +290,42 @@ def update_function(self, function: Function, code_package: Benchmark):
         name = function.name
         code_size = code_package.code_size
         package = code_package.code_location
+        benchmark = code_package.benchmark
+
+        function_cfg = FunctionConfig.from_benchmark(code_package)
+        architecture = function_cfg.architecture.value
+
         # Run AWS update
         # AWS Lambda limit on zip deployment
         if code_size < 50 * 1024 * 1024:
             with open(package, "rb") as code_body:
-                self.client.update_function_code(FunctionName=name, ZipFile=code_body.read())
+                self.client.update_function_code(
+                    FunctionName=name,
+                    ZipFile=code_body.read(),
+                    Architectures=[self._map_architecture(architecture)],
+                )
         # Upload code package to S3, then update
         else:
             code_package_name = os.path.basename(package)
+
             storage = cast(S3, self.get_storage())
             bucket = function.code_bucket(code_package.benchmark, storage)
-            storage.upload(bucket, package, code_package_name)
+            code_prefix = os.path.join(benchmark, architecture, code_package_name)
+            storage.upload(bucket, package, code_prefix)
+
             self.client.update_function_code(
-                FunctionName=name, S3Bucket=bucket, S3Key=code_package_name
+                FunctionName=name,
+                S3Bucket=bucket,
+                S3Key=code_prefix,
+                Architectures=[self._map_architecture(architecture)],
             )
         self.wait_function_updated(function)
         self.logging.info(f"Updated code of {name} function. ")
         # and update config
         self.client.update_function_configuration(
-            FunctionName=name, Timeout=function.config.timeout, MemorySize=function.config.memory
+            FunctionName=name,
+            Timeout=function.config.timeout,
+            MemorySize=function.config.memory,
         )
         self.wait_function_updated(function)
         self.logging.info(f"Updated configuration of {name} function. ")
@@ -321,8 +345,11 @@ def update_function_configuration(self, function: Function, benchmark: Benchmark
     @staticmethod
     def default_function_name(code_package: Benchmark) -> str:
         # Create function name
-        func_name = "{}-{}-{}".format(
-            code_package.benchmark, code_package.language_name, code_package.language_version
+        func_name = "{}-{}-{}-{}".format(
+            code_package.benchmark,
+            code_package.language_name,
+            code_package.language_version,
+            code_package.architecture,
         )
         return AWS.format_function_name(func_name)