diff --git a/splash/benchmark/README.rst b/splash/benchmark/README.rst
new file mode 100644
index 000000000..39a70688e
--- /dev/null
+++ b/splash/benchmark/README.rst
@@ -0,0 +1,9 @@
+This directory contains a preliminary version of splash benchmark suite.
+
+To use it, do the following:
+
+- install ``httrack``
+- run ``python download_sites.py``, it will create ``sites`` subdirectory in
+  current directory and download sites to be used in the benchmark there
+- run ``python benchmark.py`` to run the benchmark
+
diff --git a/splash/benchmark/__init__.py b/splash/benchmark/__init__.py
new file mode 100644
index 000000000..e69de29bb
diff --git a/splash/benchmark/benchmark.py b/splash/benchmark/benchmark.py
new file mode 100755
index 000000000..ca22db518
--- /dev/null
+++ b/splash/benchmark/benchmark.py
@@ -0,0 +1,264 @@
+#!/usr/bin/env python
+
+"""
+Splash benchmark script.
+
+It takes a directory downloaded with splash & httrack, fires up a static file
+server and runs a series of requests via splash on those downloaded pages.
+
+"""
+
+import json
+import logging
+import os
+import random
+from argparse import ArgumentDefaultsHelpFormatter, ArgumentParser, FileType
+from glob import glob
+from multiprocessing.pool import ThreadPool
+from pprint import pformat
+from time import time
+import re
+import sys
+
+import requests
+
+
+def make_render_png_req(splash, params):
+    """Make PNG render request via render.png endpoint."""
+    return {'url': splash.url('render.png'),
+            'params': params}
+
+
+def make_render_json_req(splash, params):
+    """Make PNG render request via JSON endpoint."""
+    json_params = params.copy()
+    json_params['png'] = 1
+    return {'url': splash.url('render.json'),
+            'params': json_params}
+
+
+def make_render_png_lua_req(splash, params):
+    """Make PNG render request via Lua execute endpoint."""
+    lua_params = params.copy()
+    lua_params['lua_source'] = """
+function main(splash)
+  assert(splash:go(splash.args.url))
+  if splash.args.wait then
+    assert(splash:wait(splash.args.wait))
+  end
+  splash:set_result_content_type("image/png")
+  return splash:png{width=splash.args.width,
+                    height=splash.args.height,
+                    render_all=splash.args.render_all}
+end
+"""
+    return {'url': splash.url('execute'),
+            'params': lua_params}
+
+
+def make_render_html_req(splash, params):
+    """Make HTML render request via render.html endpoint."""
+    return {'url': splash.url('render.html'),
+            'params': params}
+
+
+def make_render_html_json_req(splash, params):
+    """Make HTML render request via JSON endpoint."""
+    json_params = params.copy()
+    json_params['html'] = 1
+    return {'url': splash.url('render.json'),
+            'params': json_params}
+
+
+def make_render_html_lua_req(splash, params):
+    """Make HTML render request via Lua execute endpoint."""
+    lua_params = params.copy()
+    lua_params['lua_source'] = """
+function main(splash)
+  assert(splash:go(splash.args.url))
+  if splash.args.wait then
+    assert(splash:wait(splash.args.wait))
+  end
+  splash:set_result_content_type("text/html; charset=UTF-8")
+  return splash:html{}
+end
+"""
+    return {'url': splash.url('execute'),
+            'params': lua_params}
+
+
+#: Same resource may be rendered by various endpoints with slightly varying
+#: parameter combinations.  Request factories set those combinations up.
+REQ_FACTORIES = {
+    'png': [
+        make_render_png_req,
+        make_render_json_req,
+        make_render_png_lua_req,
+    ],
+    'html': [
+        make_render_html_req,
+        make_render_html_json_req,
+        make_render_html_lua_req,
+    ],
+}
+
+
+#: Port at which static pages will be served.
+PORT = 8806
+#: Combinations of width & height to test.
+WIDTH_HEIGHT = [(None, None), (500, None), (None, 500), (500, 500)]
+#: Splash & fileserver log filenames (set to None to put it to stderr).
+SPLASH_LOG = 'splash.log'
+FILESERVER_LOG = 'fileserver.log'
+#: This script is used to collect maxrss & cpu time from splash process.
+GET_PERF_STATS_SCRIPT = """
+function main(splash)
+  return splash:get_perf_stats()
+end
+"""
+
+
+parser = ArgumentParser(description=__doc__,
+                        formatter_class=ArgumentDefaultsHelpFormatter)
+parser.add_argument('--seed', type=int, default=1234, help='PRNG seed number')
+parser.add_argument('--thread-count', type=int, default=1,
+                    help='Request thread count')
+parser.add_argument('--request-count', type=int, default=10,
+                    help='Benchmark request count')
+parser.add_argument('--sites-dir', type=str, default='sites', required=True,
+                    help='Directory with downloaded sites')
+parser.add_argument('--file-server', metavar='HOST:PORT',
+                    help='Use existing file server instance available at HOST:PORT')
+parser.add_argument('--splash-server', metavar='HOST:PORT',
+                    help='Use existing Splash instance available at HOST:PORT')
+parser.add_argument('--out-file', type=FileType(mode='w'), default=sys.stdout,
+                    help='Write detailed request information in this file')
+parser.add_argument('--render-type', choices=('html', 'png'), default='png',
+                    help=('Type of rendering to benchmark'
+                          ' (either "html" or "png")'))
+
+
+def generate_requests(splash, file_server, args):
+    log = logging.getLogger('generate_requests')
+    log.info("Using pRNG seed: %s", args.seed)
+
+    # Static pages (relative to sites_dir) to be used in the benchmark.
+    log.info("sites dir: %s", args.sites_dir)
+    sites_found = glob(os.path.join(args.sites_dir, 'localhost_8806', '*.html'))
+    log.info("sites found: %s", sites_found)
+    pages = [re.sub('^%s/' % args.sites_dir.rstrip('/'), '', v) for v in sites_found]
+    for p in pages:
+        log.info("Using page for benchmark: %s", p)
+
+    request_factories = REQ_FACTORIES[args.render_type]
+
+    rng = random.Random(args.seed)
+    for i in xrange(args.request_count):
+        page = rng.choice(pages)
+        width, height = rng.choice(WIDTH_HEIGHT)
+        req_factory = rng.choice(request_factories)
+        url = file_server.url(page)
+        params = {'url': url, 'render_all': 1, 'wait': 0.1,
+                  'width': width, 'height': height}
+        log.debug("Req factory: %s, params: %s", req_factory, params)
+        yield (i + 1, args.request_count, req_factory(splash, params))
+
+
+def parallel_map(func, iterable, thread_count):
+    if thread_count == 1:
+        return map(func, iterable)
+    else:
+        pool = ThreadPool(thread_count)
+        return pool.map(func, iterable)
+
+
+def invoke_request(invoke_args):
+    log = logging.getLogger('bench_worker')
+    req_no, total_reqs, kwargs = invoke_args
+    log.info("Initiating request %d/%d: %s", req_no, total_reqs, kwargs)
+    stime = time()
+    response = requests.get(**kwargs)
+    etime = time()
+    if response.status_code != 200:
+        log.error("Non-OK response:\n%s", response.text)
+    return {'start_time': stime,
+            'end_time': etime,
+            'duration': etime - stime,
+            'endpoint': kwargs['url'],
+            'status': response.status_code,
+            'site': kwargs['params']['url'],
+            'width': kwargs['params']['width'],
+            'height': kwargs['params']['height']}
+
+
+class ExistingServerWrapper(object):
+    """Wrapper for pre-existing Splash instance."""
+    def __init__(self, server):
+        self.server = server
+        if not self.server.startswith('http://'):
+            self.server = 'http://' + self.server
+
+    def url(self, endpoint):
+        return self.server + '/' + endpoint
+
+    def __enter__(self):
+        return self
+
+    def __exit__(self, *args):
+        pass
+
+
+def main():
+    log = logging.getLogger("benchmark")
+    args = parser.parse_args()
+    (logging.getLogger('requests.packages.urllib3.connectionpool')
+     .setLevel(logging.WARNING))
+    logging.basicConfig(level=logging.DEBUG)
+
+    if args.splash_server:
+        splash = ExistingServerWrapper(args.splash_server)
+    else:
+        from splash.tests.utils import SplashServer
+        splash = SplashServer(
+            logfile=SPLASH_LOG,
+            extra_args=['--disable-lua-sandbox',
+                        '--disable-xvfb',
+                        '--max-timeout=600'])
+
+    if args.file_server:
+        file_server = ExistingServerWrapper(args.file_server)
+    else:
+        from splash.benchmark.file_server import FileServerSubprocess
+        file_server = FileServerSubprocess(port=PORT,
+                                           path=args.sites_dir,
+                                           logfile=FILESERVER_LOG)
+
+    with splash, file_server:
+        log.info("Servers are up, starting benchmark...")
+        start_res = requests.get(
+            splash.url('execute'),
+            params={'lua_source': GET_PERF_STATS_SCRIPT}).json()
+        start_time = time()
+        results = parallel_map(invoke_request,
+                               generate_requests(splash, file_server, args),
+                               args.thread_count)
+        end_time = time()
+        end_res = requests.get(
+            splash.url('execute'),
+            params={'lua_source': GET_PERF_STATS_SCRIPT}).json()
+
+    log.info("Writing stats to %s", args.out_file.name)
+    args.out_file.write(json.dumps(
+        {'maxrss': end_res['maxrss'],
+         'cputime': end_res['cputime'] - start_res['cputime'],
+         'walltime': end_time - start_time,
+         'requests': results},
+        indent=2))
+    log.info("Splash max RSS: %s B", end_res['maxrss'])
+    log.info("Splash CPU time elapsed: %.2f sec",
+             end_res['cputime'] - start_res['cputime'])
+    log.info("Wallclock time elapsed: %.2f sec", end_time - start_time)
+
+
+if __name__ == '__main__':
+    main()
diff --git a/splash/benchmark/download_sites.py b/splash/benchmark/download_sites.py
new file mode 100755
index 000000000..07a9577de
--- /dev/null
+++ b/splash/benchmark/download_sites.py
@@ -0,0 +1,119 @@
+#!/usr/bin/env python
+
+"""
+Site downloader script for Splash benchmark suite.
+"""
+
+from argparse import ArgumentParser, ArgumentDefaultsHelpFormatter
+import errno
+import json
+import os
+import re
+import subprocess
+import logging
+from urlparse import urlsplit
+
+from lxml import html
+
+import w3lib.html
+from splash.benchmark.file_server import FileServerSubprocess
+from splash.tests.stress import lua_runonce
+
+SCRIPT_HTML = """
+function main(splash)
+splash:set_images_enabled(false)
+splash:go(splash.args.url)
+splash:wait(0.5)
+return {url=splash:url(), html=splash:html()}
+end
+"""
+
+#: This UA is used by httrack to mimic Splash requests when downloading sites.
+USERAGENT = "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/534.34 (KHTML, like Gecko) Qt/4.8.1 Safari/534.34"
+
+PORT = 8806
+
+parser = ArgumentParser(description=__doc__,
+                        formatter_class=ArgumentDefaultsHelpFormatter)
+parser.add_argument('--sites-dir', default='sites',
+                    help='Directory for downloaded sites')
+
+
+def preprocess_main_page(sites_dir, url):
+    """
+    This function does several things:
+    - strip javascript so that downloaded pages look exactly the same
+    - add baseurl to resolve relative links properly (if it is missing)
+    - add meta charset description (if it is missing)
+    """
+    out = json.loads(lua_runonce(SCRIPT_HTML, url=url,
+                                 splash_args=['--disable-lua-sandbox',
+                                              '--disable-xvfb',
+                                              '--max-timeout=600'],
+                                 timeout=600.,))
+    final_url = urlsplit(out['url'])._replace(query='', fragment='').geturl()
+    # Ensure there are no scripts to be executed.
+    out['html'] = w3lib.html.remove_tags_with_content(out['html'], ('script',))
+    root = html.fromstring(out['html'], parser=html.HTMLParser(),
+                           base_url=final_url)
+    try:
+        head = root.xpath('./head')[0]
+    except IndexError:
+        head = html.Element('head')
+        root.insert(0, head)
+    if not head.xpath('./base/@href'):
+        head.insert(0, html.Element('base', {'href': final_url}))
+    if not head.xpath('./meta/@charset'):
+        head.insert(0, html.Element('meta', {'charset': 'utf-8'}))
+    out['html'] = html.tostring(root, encoding='utf-8',
+                                doctype='<!DOCTYPE html>')
+    filename = re.sub(r'[^\w]+', '_', url) + '.html'
+    with open(os.path.join(sites_dir, filename), 'w') as f:
+        f.write(out['html'])
+    return filename
+
+
+def download_sites(sites_dir, sites):
+    local_files = [preprocess_main_page(sites_dir, s) for s in sites]
+
+    local_urls = [
+        'http://localhost:%(port)d/%(filename)s' % {
+            'port': PORT, 'filename': filename
+        }
+        for filename in local_files
+    ]
+    args = ['--continue',
+            '--near',           # Fetch referred non-html files.
+            '-%P',              # Try parsing links in non-href/src sections
+            '-F', USERAGENT,    # Emulate splash UA
+            '--depth=1']
+    subprocess.check_call(['httrack'] + args + local_urls, cwd=sites_dir)
+
+
+def main():
+    args = parser.parse_args()
+    (logging.getLogger('requests.packages.urllib3.connectionpool')
+     .setLevel(logging.WARNING))
+    logging.basicConfig(level=logging.DEBUG)
+    logging.info("Starting site download suite")
+    try:
+        os.makedirs(args.sites_dir)
+    except OSError as e:
+        if e.errno != errno.EEXIST:
+            raise
+        elif not os.path.isdir(args.sites_dir):
+            raise RuntimeError("Not a directory: %s" % args.sites_dir)
+    with FileServerSubprocess(port=PORT, path=args.sites_dir):
+        download_sites(args.sites_dir, [
+            'http://www.wikipedia.org',
+            'http://www.google.com',
+            'http://www.reddit.com',
+            "http://w3.org",
+            "http://w3.org/TR/2010/REC-xhtml-basic-20101123/",
+            # "http://blog.pinterest.com",
+            # "http://imgur.com",
+        ])
+
+
+if __name__ == '__main__':
+    main()
diff --git a/splash/benchmark/file_server.py b/splash/benchmark/file_server.py
new file mode 100755
index 000000000..bb0549ce8
--- /dev/null
+++ b/splash/benchmark/file_server.py
@@ -0,0 +1,86 @@
+#!/usr/bin/env python
+
+"""Simple static file server."""
+
+import argparse
+import os
+import subprocess
+import time
+import sys
+import logging
+from contextlib import contextmanager
+
+from twisted.internet import reactor
+from twisted.web.server import Site
+from twisted.web.static import File
+from twisted.python.log import startLogging
+
+import requests
+
+parser = argparse.ArgumentParser("")
+parser.add_argument('--port', type=int, default=8806)
+parser.add_argument('--path', help='Path to be served', default='.')
+parser.add_argument('--logfile', default=sys.stderr,
+                    type=argparse.FileType(mode='w'),
+                    help='File to write logs to')
+
+
+class FileServerSubprocess(object):
+    logger = logging.getLogger('file_server')
+
+    """Serve files from specified directory statically in a subprocess."""
+    def __init__(self, port, path, logfile=None):
+        self.port = port
+        self.path = path
+        self.logfile = logfile
+        self.server = 'http://localhost:%d' % port
+
+    def url(self, endpoint):
+        return self.server + '/' + endpoint
+
+    def __enter__(self):
+        # command = ['twistd',
+        #            '-n',    # don't daemonize
+        #            'web',   # start web component
+        #            '--port', str(int(port)),
+        #            '--path', os.path.abspath(directory), ]
+        # if logfile is not None:
+        #     command += ['--logfile', logfile]
+        command = ['python', __file__,
+                   '--port', str(int(self.port)),
+                   '--path', os.path.abspath(self.path)]
+        if self.logfile is not None:
+            command += ['--logfile', self.logfile]
+        self.logger.info("Starting file server subprocess: %s", command)
+        self._site_server = subprocess.Popen(command)
+        # It might take some time to bring up the server, wait for up to 10s.
+        for i in xrange(100):
+            try:
+                self.logger.info("Checking if file server is active")
+                requests.get(self.url(''))
+                break
+            except requests.ConnectionError:
+                time.sleep(0.1)
+        else:
+            msg = "File server subprocess startup timed out"
+            if self.logfile:
+                with open(self.logfile, 'r') as log_f:
+                    msg += ", logs:\n" + log_f.read()
+            raise RuntimeError(msg)
+
+    def __exit__(self, *args):
+        self._site_server.kill()
+        self._site_server.wait()
+
+
+def main():
+    args = parser.parse_args()
+    startLogging(args.logfile)
+    resource = File(os.path.abspath(args.path))
+    site = Site(resource)
+    reactor.listenTCP(args.port, site)
+    reactor.run()
+
+
+if __name__ == '__main__':
+    main()
diff --git a/splash/conftest.py b/splash/conftest.py
index 2a20cc38a..f93ed2990 100644
--- a/splash/conftest.py
+++ b/splash/conftest.py
@@ -1,6 +1,7 @@
 # -*- coding: utf-8 -*-
 from __future__ import absolute_import
 from splash import lua
+import glob
 
 collect_ignore = []
 
@@ -15,3 +16,8 @@
         'kernel/__main__.py',
         'kernel/__init__.py',
     ]
+
+collect_ignore.extend([
+    'benchmark/download_sites.py',
+    'benchmark/file_server.py',
+    'benchmark/benchmark.py'])
diff --git a/splash/qtrender_lua.py b/splash/qtrender_lua.py
index 82f6f8fe5..0e12b8503 100644
--- a/splash/qtrender_lua.py
+++ b/splash/qtrender_lua.py
@@ -497,7 +497,7 @@ def get_perf_stats(self):
         rss_mul = 1 if sys.platform == 'darwin' else 1024
         return {'maxrss': rusage.ru_maxrss * rss_mul,
                 'cputime': rusage.ru_utime + rusage.ru_stime,
-                'walltime': time.time()}
+                 'walltime': time.time()}
 
     def get_real_exception(self):
         if self._exceptions: