On Fri, 29 Jul 2016 17:10:39 +0800 [email protected] wrote: > Hi, Micheal and jyh, > > It's great that you working on this. > > Hugepage, Numa, Cpu pinning are common requirements in NFV > application, which will be consumed by KVM, OVS and FD.IO projects in > OPNFV. For a VNF, it also has the same requirements even without KVM, > OVS and FD.IO enhancement such as used in SR-IOV. > > In OPNV, currently a scenario is defined by: > Installer(4)*sdntype(nosdn/odl_l2/odl_l3/opencontrail/onos)*feature(more > and more)*ha(noha/ha). > > There are too many scenarios now, and it will introduce tens of > scenarios if we introduce hugepage as scenario factor. It will also > cost too much work and too many hardware/CI resources. Shall we > configure this feature in current scenarios other than add new > scenarios. It can be a input parameter when triggering these > scenarios.
Hmm, how to test them is something challenge. We should analysis these features and decide one by one. For example, I'd assume huge pages and CPU pinning support should always be enabled on openstack for all scenarios and we should apply these features to VM created by yardstick for KVM4NFV/VSPERF test cases, which has much higher latency/throughput requirement. > > This is a common requirement in Genesis, and let's have more > discussion and implement it in Fuel. It will give an good example for > other installers. Will this be discussed on the Generis meeting? Will it be supported on C release or D release? Thanks --jyh > > > BR, > Julien > > > > > > > > > > Julien(Jun) Zhang > NFV Architect > > D206(D2028), R&D Building, ZTE Plaza, #889, Bibo RD, > Pudong District, Shang Hai, P.R.China, 201203 > T:+86 21 68897791 M:+86 15317680806 F:+86 21 50801070 > E: [email protected] G: [email protected] > www.zte.com.cn, www.opnfv.org > > > > > > 发件人: Michal Skalski <[email protected]> > 收件人: yunhong jiang <[email protected]>, > 抄送: "[email protected]" > <[email protected]> > 日期: 2016/07/29 15:06 > 主题: Re: [opnfv-tech-discuss] [fuel][new scenario][hugepage]Fuel > configure hugepage with CI. > 发件人: [email protected] > > > > Hi, > > Yes, fuel support such configuration, we are currently working on pod > configuration / existing scenario improvement: > > https://jira.opnfv.org/browse/PHAROS-230 > https://gerrit.opnfv.org/gerrit/#/c/17727/ > https://gerrit.opnfv.org/gerrit/#/c/16913/ > > Regards, > Michal > > On Fri, Jul 29, 2016 at 8:38 AM, yunhong jiang < > [email protected]> wrote: > On Fri, 29 Jul 2016 14:22:07 +0800 > [email protected] wrote: > > > Hi fuel team， > > > > I notice that fuel@opnfv support hugepage configuration. ZTE-POD1 > > want to configure hugepage with CI. > > Hi, Zhihui > Thanks for sharing this information. To achieve good performance > and low latency with openstack, a lot of configuration is > required, like hugue page, CPU reservation, NUMA node (I gave a > presentation on openstack summit on this). > > How can I check the fuel's support to these configuration? It will > be perfect if we can add such support to the scenario tests. > > Thanks > --jyh > _______________________________________________ > opnfv-tech-discuss mailing list > [email protected] > https://lists.opnfv.org/mailman/listinfo/opnfv-tech-discuss > _______________________________________________ > opnfv-tech-discuss mailing list > [email protected] > https://lists.opnfv.org/mailman/listinfo/opnfv-tech-discuss > > _______________________________________________ opnfv-tech-discuss mailing list [email protected] https://lists.opnfv.org/mailman/listinfo/opnfv-tech-discuss
