PS...I do use once process per core with the gevent loop. I assume that is best practice?
On Sat, Nov 10, 2012 at 2:37 AM, David Montgomery <[email protected] > wrote: > Wow..cool...I am using the graphite plugin. I see two new > metrics......busyiness and harikari. What is the best explanation of those > two? > > I am pretty much at 100% on all busyiness on 3 servers with 8 cores each. > But I am doing 6K qps. Clearly I need more servers but is rather a black > art to tell if I am truly using the servers to the max before I scale out. > I dont know:( > > > > On Sat, Nov 10, 2012 at 1:15 AM, Roberto De Ioris <[email protected]>wrote: > >> >> Il giorno 09/nov/2012, alle ore 18:05, Jeff Van Voorst < >> [email protected]> ha scritto: >> >> > The best answer is it depends on your application and usage patterns. >> You will want to test loading your system till it breaks for different >> values of n. >> > >> > Potential starting values of n might be small (e.g. 8) for heavy >> computational apps, and much larger for database query or I/O heavy apps >> (e.g. 256 or even 1024). Note that these are starting values, and the only >> way to know is to test your app as you see and expect users to use it. >> > >> > --Jeff >> > >> > >> >> A very important test is using the stats subsystem of 1.4 that exports >> information about cores (in the case of the gevent plugin, a core is mapped >> to a greenlet). >> >> As the uWSGI async mode tend to reuse the first available core, you will >> get a pretty accurate info about the need of adding more cores/greenlets >> (or to remove them if they are unused) >> >> -- >> Roberto De Ioris >> http://unbit.it >> JID: [email protected] >> >> _______________________________________________ >> uWSGI mailing list >> [email protected] >> http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi >> > >
_______________________________________________ uWSGI mailing list [email protected] http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi
