Hi James, This is awesome. I seem to have misplaced my 540-node cluster. ;-)
Is it possible for you to also patch in https://review.openstack.org/#/c/132372/ ? In my rally testing of port retrieval, this one probably made the most significant improvement. On Nov 12, 2014 9:26 AM, "James Page" <[email protected]> wrote: > -----BEGIN PGP SIGNED MESSAGE----- > Hash: SHA256 > > Hi Ihar > > On 11/11/14 19:39, Ihar Hrachyshka wrote: > >> there is a series of Neutron backports in the Juno queue that are > >> > >>> intended to significantly improve service performance when > >>> handling security groups (one of the issues that are main pain > >>> points of current users): - https://review.openstack.org/130101 > >>> - https://review.openstack.org/130098 - > >>> https://review.openstack.org/130100 - > >>> https://review.openstack.org/130097 - > >>> https://review.openstack.org/130105 The first four patches are > >>> optimizing db side (controller), while the last one is to avoid > >>> fetching security group rules by OVS agent when firewall is > >>> disabled. > > In terms of putting some figures around how these proposed stable > patches help improve a Neutron based Juno cloud, I can provide some > metrics based on recent testing that Canonical did in-conjuction with HP. > > The cloud we deployed was all based on Intel Atom Quad Core processors, > with 16G of RAM and SSD disk; 540 servers in total including 8 nova > controllers and 4 neutron controllers. OpenStack Juno release on > Ubuntu 14.04. > > With around 12,000 running instances, which was as far as I could push > a vanilla ML2/ovs based Juno cloud, the load on the 4 neutron > controllers was around 40 with CPU maxing out all of the time - which > pretty much mean't it was impossible to create any new instances due > to vif plugging timeouts in nova waiting for neutron to complete > network setup. > > I patched in: > > https://review.openstack.org/#/c/130101/ > https://review.openstack.org/#/c/130098/ > https://review.openstack.org/#/c/130100/ > https://review.openstack.org/#/c/130105/ > > and re-ran the same test; the messaging load on the RabbitMQ server at > 12,000 instances was considerably less in terms of volume, and the > load on the 4 neutron controllers was around 10 (vs 40) with CPU at > around 55->65% utilization - so still pretty busy, but a better > situation than without the patches. > > My testing was quite synthetic (boot small instances until things > start to break) but it does illustrate the difference these patches make. > > HTH > > James > > - -- > James Page > Ubuntu and Debian Developer > [email protected] > [email protected] > -----BEGIN PGP SIGNATURE----- > Version: GnuPG v1 > > iQIcBAEBCAAGBQJUY5czAAoJEL/srsug59jDIJ8QAJAK8aWUSQGyPrcAqvKi+OE7 > vS4NhVWog1ifubPcIpDstAoELHIfQVQKaryCN7oXAzQ0Yyp68DE68mw+o8rrly2S > 25gPebORNath1BOJMMlv5iRS0lVN30cfmRrs9nfQ5bdAE6qkaPlofG9GsGRggCG2 > feewRR9w+PFFQQ9NdsZ141FoQDtpLjhY095rEwzUhyah8spM2w2er2XiEJLHRTI/ > HcJybUSX/Nu8OV4FJ6dn+pebWv1iWgzNOV/eqCYHf1Mx9G6HrB8ZQpv486LznyX1 > PSNuiVMgUFcSWUcN1lFQSEe/ASW+G2t3/aEMKZBXiXsO3DTORtZ79oCTkzipkehj > 18ztLr+nkCDrdGzbvkD6LWGt9F7MjTzsXao4RwGe/EiRBvcrvnHpkc5kfaW2aIb3 > +rH8pcHpfaC04y7Zy492lFrkmrXn+73c2a+hS+gS3bMmQ1bcwF+QeeXunsMajgVo > CQW98n3HJI/jAjCBEbV5cmmw+BXQDWOHYlP+tZiAMC5Tnj42/9+K+KWZr+truhLK > cKGFlM+vaVsykAh9KIf1E/e6G72o/kihXDUnpx/mSk27sxDILEz9ItcQRJgpQCPN > cH3sIj+qG76NDqIhdLYs8LgyjwQI2SdOeSi+32oCCe3tnaI35FKKuRMI0oSP0HKn > 3U7bekTsjXhlBWusW9Wb > =WaFI > -----END PGP SIGNATURE----- > > _______________________________________________ > OpenStack-dev mailing list > [email protected] > http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev >
_______________________________________________ OpenStack-dev mailing list [email protected] http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
