« ink!More B&Ws »


Comment from: Chad Whitacre [Visitor] · http://tech.whit537.org/

Good work Robert! I upgraded Aspen trunk to the 3.0.0 server the other day ... the AI_PASSIVE changes threw me for a loop (an empty string for host now means AF_INET6 rather than AF_INET) but other than that I'm excited about the improvements. :)

BTW, I believe the problem with > 50 threads on Windows is due to a limit of 64 sockets in Windows select implementation, which is there to encourage use of some other MS-specific API. Try -c64 and -c65 to illustrate the limit.

12/26/06 @ 09:15
Comment from: Chad Whitacre [Visitor] · http://tech.whit537.org/

Sorry I missed you on IRC. Here's the reference I had in mind for the 64-socket limit:


12/27/06 @ 14:10
Comment from: fumanchu [Member] Email

Right. My IRC comment was noticing that both Python and Apache 2.0.59 set FD_SETSIZE to something larger than 64 (so I was initially confused why the limit was still in effect). However, it doesn't do it in the right place; adding it to httpd-2.0.59\srclib\apr\include\apr.hw just before the winsock2.h include allows ab to use more than 64 sockets on Windows. You also need to tell benchmark.py about the location of the recompiled ab.exe (in the AB_PATH global variable), and remove or rewrite the special-casing for the number of threads when sys.platform == win32. Here's the results for the full CherryPy stack:

Client Thread Report (1000 requests, 14 byte response body, 10 server threads):

threads | Completed | Failed | req/sec | msec/req | KB/sec |
25 | 1000 | 0 | 531.15 | 1.883 | 84.45 |
50 | 1000 | 0 | 520.08 | 1.923 | 82.69 |
100 | 1000 | 0 | 499.28 | 2.003 | 79.39 |
200 | 1000 | 0 | 480.08 | 2.083 | 76.33 |
400 | 1000 | 0 | 436.05 | 2.293 | 69.33 |
Average | 1000.0 | 0.0 | 493.328 | 2.037 | 78.438 |

12/28/06 @ 01:01
Comment from: Grumpy Old Man [Visitor]

How about defining "modest laptop", or better still, give figures for requesting a static file containing the same text from Apache when running on the same machine. Ensure that the version of Apache and which Apache MPM is being used are stated as well.

If you give the figures for Apache then people with a different machine can run a test with Apache (something most people would have access to) and by comparing their Apache results to your results, then extrapolate as to what performance CherryPy WSGI server will yield without actually having to install CherryPy. Without some sort of reference like this the results are pretty meaningless and can't easily be compared to other systems. :-)

01/22/07 @ 03:27
Comment from: Cliff Wells [Visitor] · http://pentropy.twisty-industries.com/

Actually, there's little doubt FAPWS is much faster (I've benchmarked ~5K req/s with FAPWS vs ~1K req/s for CP3's wsgiserver on the same hardware) serving a simple "hello, world." type page.

Still, I'd suggest that CP3 currently has the fastest working implementation of a WSGI server as FAPWS is highly experimental, unstable, mostly undocumented, and not really suitable for anything but benchmarks ;-)

BTW, isn't "I am a real human" just what a not-real human would say?

08/15/08 @ 04:24

Leave a comment

Your email address will not be revealed on this site.

Your URL will be displayed.

Please enter the phrase "I am a real human." in the textbox above.
(Line breaks become <br />)
(Name, email & website)
(Allow users to contact you through a message form (your email will not be revealed.)