diff --git a/azdev/commands.py b/azdev/commands.py
index c3f191998..bd8e4be7a 100644
--- a/azdev/commands.py
+++ b/azdev/commands.py
@@ -43,6 +43,7 @@ def operation_group(name):
 
     with CommandGroup(self, 'perf', operation_group('performance')) as g:
         g.command('load-times', 'check_load_time')
+        g.command('benchmark', 'benchmark', is_preview=True)
 
     with CommandGroup(self, 'extension', operation_group('extensions')) as g:
         g.command('add', 'add_extension')
diff --git a/azdev/help.py b/azdev/help.py
index c102d44d6..9b1b21c9e 100644
--- a/azdev/help.py
+++ b/azdev/help.py
@@ -144,6 +144,12 @@
     short-summary: Verify that all modules load within an acceptable timeframe.
 """
 
+helps['perf benchmark'] = """
+    short-summary: Display benchmark staticstic of Azure CLI (Extensions) commands via execute it with "python -m azure.cli {COMMAND}" in a separate process.
+    examples:
+        - name: Run benchmark on "network application-gateway" and "storage account"
+          text: azdev perf benchmark "network application-gateway -h" "storage account" "version" "group list"
+"""
 
 helps['extension'] = """
     short-summary: Control which CLI extensions are visible in the development environment.
diff --git a/azdev/operations/performance.py b/azdev/operations/performance.py
index 5cc7eec23..8dfdb2673 100644
--- a/azdev/operations/performance.py
+++ b/azdev/operations/performance.py
@@ -5,12 +5,13 @@
 # -----------------------------------------------------------------------------
 
 import re
+import timeit
 
 from knack.log import get_logger
 from knack.util import CLIError
 
 from azdev.utilities import (
-    display, heading, subheading, cmd, require_azure_cli)
+    display, heading, subheading, cmd, py_cmd, require_azure_cli)
 
 logger = get_logger(__name__)
 
@@ -106,17 +107,20 @@ def _claim_higher_threshold(val):
 in any modified files.
 """)
 
-    display('== PASSED MODULES ==')
+    display("== PASSED MODULES ==")
     display_table(passed_mods)
-    display('\nPASSED: Average load time all modules: {} ms'.format(
-        int(passed_mods[TOTAL]['average'])))
+    display(
+        "\nPASSED: Average load time all modules: {} ms".format(
+            int(passed_mods[TOTAL]["average"])
+        )
+    )
 
 
 def mean(data):
     """Return the sample arithmetic mean of data."""
     n = len(data)
     if n < 1:
-        raise ValueError('len < 1')
+        raise ValueError("len < 1")
     return sum(data) / float(n)
 
 
@@ -140,3 +144,103 @@ def display_table(data):
     for key, val in data.items():
         display('{:<20} {:>12.0f} {:>12.0f} {:>12.0f} {:>25}'.format(
             key, val['average'], val['threshold'], val['stdev'], str(val['values'])))
+
+
+# require azdev setup
+def benchmark(commands, runs=20):
+    if runs <= 0:
+        raise CLIError("Number of runs must be greater than 0.")
+
+    max_len_cmd = max(commands, key=len)
+
+    line_tmpl = "| {" + "cmd:" + "<" + str(len(max_len_cmd)) + "s} |"
+    line_tmpl = line_tmpl + " {min:10s} | {max:10s} | {avg:10s} | {mid:10s} | {std:10s} | {runs:10s} |"
+
+    line_head = line_tmpl.format(
+        cmd="Command",
+        min="Min",
+        max="Max",
+        avg="Mean",
+        mid="Median",
+        std="Std",
+        runs="Runs",
+    )
+
+    logger.warning(line_head)
+    logger.warning("-" * (85 + len(max_len_cmd)))
+
+    import multiprocessing
+
+    # Measure every wanted commands
+    for raw_command in commands:
+        logger.info("Measuring %s...", raw_command)
+
+        pool = multiprocessing.Pool(multiprocessing.cpu_count(), _benchmark_process_pool_init)
+
+        # try/except like this because of a bug of Python multiprocessing.Pool (https://bugs.python.org/issue8296)
+        # Discussion on StackOverflow:
+        # https://stackoverflow.com/questions/1408356/keyboard-interrupts-with-pythons-multiprocessing-pool/1408476
+        try:
+            time_series = pool.map_async(_benchmark_cmd_timer, [raw_command] * runs).get(1000)
+        except multiprocessing.TimeoutError:
+            pool.terminate()
+            break
+        else:
+            pool.close()
+        pool.join()
+
+        staticstic = _benchmark_cmd_staticstic(time_series)
+
+        line_body = line_tmpl.format(
+            cmd=raw_command,
+            min=str(staticstic["min"]),
+            max=str(staticstic["max"]),
+            avg=str(staticstic["avg"]),
+            mid=str(staticstic["media"]),
+            std=str(staticstic["std"]),
+            runs=str(runs),
+        )
+        logger.warning(line_body)
+
+    logger.warning("-" * (85 + len(max_len_cmd)))
+
+
+def _benchmark_process_pool_init():
+    import signal
+    signal.signal(signal.SIGINT, signal.SIG_IGN)
+
+
+def _benchmark_cmd_timer(raw_command):
+    s = timeit.default_timer()
+    py_cmd("azure.cli {}".format(raw_command), is_module=True)
+    e = timeit.default_timer()
+    return round(e - s, 4)
+
+
+def _benchmark_cmd_staticstic(time_series: list):
+    from math import sqrt
+
+    time_series.sort()
+
+    size = len(time_series)
+
+    if size % 2 == 0:
+        mid_time = (time_series[size // 2 - 1] + time_series[size // 2]) / 2
+    else:
+        mid_time = time_series[(size - 1) // 2]
+
+    min_time = time_series[0]
+    max_time = time_series[-1]
+    avg_time = sum(time_series) / size
+
+    std_deviation = sqrt(
+        sum([(t - avg_time) * (t - avg_time) for t in time_series]) / size
+    )
+
+    return {
+        "min": round(min_time, 4),
+        "max": round(max_time, 4),
+        "media": round(mid_time, 4),
+        "avg": round(avg_time, 4),
+        "std": round(std_deviation, 4),
+    }
diff --git a/azdev/params.py b/azdev/params.py
index 39982be85..a44fae60c 100644
--- a/azdev/params.py
+++ b/azdev/params.py
@@ -101,6 +101,10 @@ def load_arguments(self, _):
     with ArgumentsContext(self, 'perf') as c:
         c.argument('runs', type=int, help='Number of runs to average performance over.')
 
+    with ArgumentsContext(self, 'perf benchmark') as c:
+        c.positional('commands', nargs="+", help="Command prefix to run benchmark")
+        c.argument('top', type=int, help='Show N slowest commands. 0 for all.')
+
     with ArgumentsContext(self, 'extension') as c:
         c.argument('dist_dir', help='Name of a directory in which to save the resulting WHL files.')
 
diff --git a/azure-pipelines.yml b/azure-pipelines.yml
index ab41a87cb..3c8ad283b 100644
--- a/azure-pipelines.yml
+++ b/azure-pipelines.yml
@@ -41,7 +41,7 @@ jobs:
     inputs:
       targetType: 'filePath'
       filePath: scripts/ci/run_tox.sh
-      
+
 - job: Tox38
   displayName: 'Tox: Python 3.8'
   condition: succeeded()
@@ -267,3 +267,81 @@ jobs:
       # verify azdev style works
       azdev style redis
     displayName: 'Test azdev style'
+
+- job: PerformanceCheckOnLinux
+  displayName: "Performance Check on Linux"
+  dependsOn: BuildPythonWheel
+  pool:
+    vmImage: 'ubuntu-16.04'
+  strategy:
+    matrix:
+      Python36:
+        python.version: '3.6'
+      Python38:
+        python.version: '3.8'
+  steps:
+    - task: DownloadPipelineArtifact@1
+      displayName: 'Download Build'
+      inputs:
+        TargetPath: '$(Build.ArtifactStagingDirectory)/pypi'
+        artifactName: pypi
+    - task: UsePythonVersion@0
+      displayName: 'Use Python $(python.version)'
+      inputs:
+        versionSpec: '$(python.version)'
+    - bash: |
+        set -ev
+        . scripts/ci/install.sh
+        azdev --version
+      displayName: 'Azdev Setup'
+    - bash: |
+        set -ev
+        . env/bin/activate
+        azdev perf load-times
+      displayName: "Load Performance"
+    - bash: |
+        set -ev
+        . env/bin/activate
+        azdev perf benchmark "version" "network vnet -h" "rest -h" "storage account"
+      displayName: "Execution Performance"
+
+# - job: PerformanceCheckOnWindows
+#   displayName: "Performance Check on Windows"
+#   dependsOn: BuildPythonWheel
+#   pool:
+#     vmImage: 'vs2017-win2016'
+#   strategy:
+#     matrix:
+#       Python36:
+#         python.version: '3.6'
+#       Python38:
+#         python.version: '3.8'
+#   steps:
+#     - task: DownloadPipelineArtifact@1
+#       displayName: 'Download Build'
+#       inputs:
+#         TargetPath: '$(Build.ArtifactStagingDirectory)/pypi'
+#         artifactName: pypi
+#     - task: UsePythonVersion@0
+#       displayName: 'Use Python $(python.version)'
+#       inputs:
+#         versionSpec: '$(python.version)'
+#     - powershell: |
+#         python -m venv env
+#         .\env\Scripts\Activate.ps1
+#         pip install --user -U pip setuptools wheel -q
+#         pip install --user $(find ${BUILD_ARTIFACTSTAGINGDIRECTORY}/pypi -name *.tar.gz) -q
+#         git clone https://github.com/Azure/azure-cli.git
+#         git clone https://github.com/Azure/azure-cli-extensions.git
+#         azdev setup -c -r azure-cli-extensions
+
+#         azdev --version
+#       displayName: 'Azdev Setup'
+#     - powershell: |
+#         .\env\Scripts\Activate.ps1
+#         azdev perf load-times
+#       displayName: "Load Performance"
+#     - powershell: |
+#         .\env\Scripts\Activate.ps1
+#         azdev perf benchmark "version" "network vnet -h" "rest -h" "storage account"
+#       displayName: "Execution Performance"