What mod_wsgi settings are you using? I tend to use modwsgi daemon mode and at least 1 process per CPU, and then keep adding threads to the point where we have one thread per apache process, or one thread per allowed database connection. But slow clients, database hit free pages, and other factors can impact the ratios a lot, so I'd do a couple of tests and figure out what settings work best for you.
On a work project of mine we are actually running modwsgi behing nginx and serving up all the static content from nginx. turbogears.org currently serves static files via apache, on a shared host, so it's not at all a fair comparison ;) On Tue, May 19, 2009 at 12:45 PM, cd34 <[email protected]> wrote: > > I am working on a turbogears app and decided to test the different > methods of deploying an app and I'm a little concerned. Since this > app requires SSL, and paster doesn't handle SSL, what are the > suggested methods for deployment? > > Paster gives reasonably good performance during development for a > single user, however, I cannot use SSL with Paster. > > mod_wsgi appears to work ok until there are multiple requests coming > in and response times on reloaded pages falls off considerably. I > don't believe I would have too many issues with this application, but, > a reasonably successful site would appear to have issues even without > SSL. > > How is turbogears.org set up? mod_wsgi behind mpm-prefork? or mpm- > worker? > > apachebench on a very simple test, hitting a page that requires > authentication, 100 requests, no keepalives, no concurrent connections > comes up with: > > paster: Requests per second: Requests per second: 76.32 [#/sec] > (mean) > apache2-mpm-prefork Requests per second: 57.38 [#/sec] (mean) > apache2-mpm-worker Requests per second: 89.98 [#/sec] (mean) > > As a baseline that is relatively decent. The current php application > we're running is at: Requests per second: 72.51 [#/sec] (mean) on > lesser hardware. > > If we bump up the concurrency, we start to run into issues: > > At 10 concurrent requests, we get the following: > > paster: Requests per second: 52.94 [#/sec] (mean) > apache2-mpm-prefork: Requests per second: 54.45 [#/sec] (mean) > apache2-mpm-worker: Requests per second: 53.61 [#/sec] (mean) > > whereas our php application with 10 concurrent sessions is at: > Requests per second: 62.35 [#/sec] (mean). > > If we test 100 concurrent requests, mpm-worker really shines through > by staying at roughly Requests per second: 53.46 [#/sec] (mean) and > paster comes through with Requests per second: 54.85 [#/sec] (mean) > > These results don't come close to the Requests per second: 86.33 [#/ > sec] (mean) at http://turbogears.org/2.0/ Since you have ServerTokens > set to prod, I couldn't really guess how you had deployed the site. > Since you use Discus, I'm assuming Identity is turned off. > > So even with slight concurrency, mod_wsgi (or PasteWSGIServer/0.5) is > taking a rather huge hit. Unless I have missed something obvious with > the testing, writing a scaleable application with TurboGears is going > to require considerable server resources. > > A quick test to check mod_wsgi performance on my setup resulted in: > > Requests per second: 2370.11 [#/sec] (mean) > > Concurrency set to 20, 1000 requests. Even at 50 or 100 concurrent > requests, the numbers don't change that dramatically. > > This seems to suggest the load isn't from mod_wsgi, but from TG2. > > What have I done differently than the Turbogears site? Even without > authentication, with 10 concurrent requests I cannot get my > installation anywhere near 86.33 requests per second even on fairly > powerful hardware (X3220 @ 2.40GHz, quadcore, 8gb ram) The only way > I could come close to the 86.33 was to run the app on a 2 server > cluster. > > I have yet to try nginx/mod_wsgi, but, I believe my results might be > similar. Apache was not configured to serve static images. Since > apachebench only pulled a single page without any stylesheets/images, > the test should mimic a properly configured setup by testing only the > response time that turbogears took to check to see if the current > session was authenticated and presented a login page. > > What deployment options are there? > > paster > proxy -> paster > apache2-mpm-prefork -> mod_wsgi > apache2-mpm-worker -> mod_wsgi > nginx -> mod_wsgi > apache2 -> mod_python - run the entire virtualenvironment within > mod_python? > > Based on the numbers above, apache2-mpm-worker appears to give me the > flexibility to allow TG2 to handle only the pages it needs while > apache serves the static content. > > For a site that will see 100k unique visitors with 4-5 pageviews a > person a day, what bottlenecks should I expect? > > The site shootq.com which is linked from the frontpage of > turbogears.org/2.0 as a site using TG2 in production appears to be > using CakePHP as their framework. Does anyone use TurboGears on a > forward-facing website? What issues have you run into? How did you > deploy and what type of traffic are you seeing? > > > -- Mark Ramm-Christensen email: mark at compoundthinking dot com blog: www.compoundthinking.com/blog --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "TurboGears" group. To post to this group, send email to [email protected] To unsubscribe from this group, send email to [email protected] For more options, visit this group at http://groups.google.com/group/turbogears?hl=en -~----------~----~----~----~------~----~------~--~---
