import socket
import signal
import pyev

class Client(object):
    def start(self, loop, sock):
        self.msg = "HTTP/1.0 200 OK\r\nContent-Length: 5\r\n\r\nPong!\r\n"
        self.sock = sock
        self.write = self.write_msg
        self.writewatcher = pyev.Io(self.sock, pyev.EV_WRITE, loop, self.write)
        self.writewatcher.start()

    def write_msg(self, watcher, events):
        self.sock.send(self.msg)
        self.writewatcher.stop()
        self.sock.close()

class Server(object):
    def start(self, loop):
        self.sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
        self.sock.bind(('localhost', 10000))
        self.sock.listen(500)
        self.acceptwatcher = pyev.Io(self.sock, pyev.EV_READ, loop, self.accept)
        self.acceptwatcher.start()

    def accept(self, watcher, events):
        (sock, addr) = self.sock.accept()
        client = Client()
        client.start(watcher.loop, sock)

def interrupt(watcher, events):
    watcher.loop.unloop()

loop = pyev.default_loop()
sigwatch = pyev.Signal(signal.SIGINT, loop, interrupt)
sigwatch.start()
srv = Server()
srv.start(loop)
loop.loop()

Here is the results using ra=10000

httperf –hog –server localhost –num-conn 40000 –port 10000 –ra 10000 –timeout 5
httperf –hog –timeout=5 –client=0/1 –server=localhost –port=10000 –uri=/ –rate=10000 –send-buffer=4096 –recv-buffer=16384 –num-conns=40000 –num-calls=1
Maximum connect burst length: 41

Total: connections 40000 requests 40000 replies 40000 test-duration 3.997 s

Connection rate: 10008.5 conn/s (0.1 ms/conn, <=57 concurrent connections)
Connection time [ms]: min 0.1 avg 0.3 max 4.5 median 0.5 stddev 0.2
Connection time [ms]: connect 0.1
Connection length [replies/conn]: 1.000

Request rate: 10008.5 req/s (0.1 ms/req)
Request size [B]: 62.0

Reply rate [replies/s]: min 0.0 avg 0.0 max 0.0 stddev 0.0 (0 samples)
Reply time [ms]: response 0.2 transfer 0.0
Reply size [B]: header 38.0 content 5.0 footer 0.0 (total 43.0)
Reply status: 1xx=0 2xx=40000 3xx=0 4xx=0 5xx=0

CPU time [s]: user 1.19 system 2.79 (user 29.9% system 69.9% total 99.7%)
Net I/O: 1026.3 KB/s (8.4*10^6 bps)

Errors: total 0 client-timo 0 socket-timo 0 connrefused 0 connreset 0
Errors: fd-unavail 0 addrunavail 0 ftab-full 0 other 0

Note that the sockets are in fact blocking, I never set nonblocking, I have no doubt I could do even better with nonblocking.

If this bench is saying anything, its saying nothing comes even close to pyev in the python world, nothing.

Just to really push things, I tried 20k per second…

httperf –hog –server localhost –num-conn 40000 –port 10000 –ra 20000 –timeout 5
httperf –hog –timeout=5 –client=0/1 –server=localhost –port=10000 –uri=/ –rate=20000 –send-buffer=4096 –recv-buffer=16384 –num-conns=40000 –num-calls=1
Maximum connect burst length: 33

Total: connections 40000 requests 40000 replies 40000 test-duration 2.004 s

Connection rate: 19963.5 conn/s (0.1 ms/conn, <=56 concurrent connections)
Connection time [ms]: min 0.1 avg 0.5 max 2.7 median 0.5 stddev 0.2
Connection time [ms]: connect 0.3
Connection length [replies/conn]: 1.000

Request rate: 19963.5 req/s (0.1 ms/req)
Request size [B]: 62.0

Reply rate [replies/s]: min 0.0 avg 0.0 max 0.0 stddev 0.0 (0 samples)
Reply time [ms]: response 0.3 transfer 0.0
Reply size [B]: header 38.0 content 5.0 footer 0.0 (total 43.0)
Reply status: 1xx=0 2xx=40000 3xx=0 4xx=0 5xx=0

CPU time [s]: user 0.35 system 1.64 (user 17.6% system 81.8% total 99.5%)
Net I/O: 2047.0 KB/s (16.8*10^6 bps)

Errors: total 0 client-timo 0 socket-timo 0 connrefused 0 connreset 0
Errors: fd-unavail 0 addrunavail 0 ftab-full 0 other 0

And finally how the twisted example does on my machine in comparison, exactly as taken from this blog when given a rate of 10k…

httperf –hog –server localhost –num-conn 40000 –port 10000 –ra 10000 –timeout 5
httperf –hog –timeout=5 –client=0/1 –server=localhost –port=10000 –uri=/ –rate=10000 –send-buffer=4096 –recv-buffer=16384 –num-conns=40000 –num-calls=1
Maximum connect burst length: 16

Total: connections 26269 requests 26269 replies 26269 test-duration 6.261 s

Connection rate: 4195.4 conn/s (0.2 ms/conn, <=1022 concurrent connections)
Connection time [ms]: min 0.5 avg 152.4 max 3162.8 median 116.5 stddev 305.9
Connection time [ms]: connect 31.7
Connection length [replies/conn]: 1.000

Request rate: 4195.4 req/s (0.2 ms/req)
Request size [B]: 62.0

Reply rate [replies/s]: min 5247.1 avg 5247.1 max 5247.1 stddev 0.0 (1 samples)
Reply time [ms]: response 120.7 transfer 0.0
Reply size [B]: header 38.0 content 5.0 footer 0.0 (total 43.0)
Reply status: 1xx=0 2xx=26269 3xx=0 4xx=0 5xx=0

CPU time [s]: user 0.24 system 6.00 (user 3.9% system 95.9% total 99.8%)
Net I/O: 430.2 KB/s (3.5*10^6 bps)

Errors: total 13731 client-timo 0 socket-timo 0 connrefused 0 connreset 0
Errors: fd-unavail 13731 addrunavail 0 ftab-full 0 other 0

It basically fails to even finish…

FYI Syncess, a high performance asynchronous client and server library using Stackless Python has born after this comparison. See Syncless at http://code.google.com/p/syncless/

I’ve also written a feature comparison of event-driven and coroutine-based non-blocking I/O libraries for Python. Read it at: http://ptspts.blogspot.com/2010/05/feature-comparison-of-python-non.html

Kamaelia project is missing.

The current Medusa is at http://www.amk.ca/python/code/medusa.html.