Hi team, I have reinstalled StarlingX on my 4 bare metals but meet some problems. As soon as I deploy StarlingX successfully, I will try to add property “hw:mem_page_size=large” to flavor and then create VMs. Best Regards, Xu, Chenjie From: Peters, Matt [mailto:Matt.Peters@windriver.com] Sent: Thursday, April 4, 2019 3:21 AM To: Xu, Chenjie <chenjie.xu@intel.com>; Liu, ZhipengS <zhipengs.liu@intel.com>; He, Yongli <yongli.he@intel.com> Cc: 'starlingx-discuss@lists.starlingx.io' <starlingx-discuss@lists.starlingx.io>; Khalil, Ghada <Ghada.Khalil@windriver.com>; Zhao, Forrest <forrest.zhao@intel.com>; Rowsell, Brent <Brent.Rowsell@windriver.com>; Gauld, James <James.Gauld@windriver.com>; Le, Huifeng <huifeng.le@intel.com>; Martinez Monroy, Elio <elio.martinez.monroy@intel.com>; Perez, Ricardo O <ricardo.o.perez@intel.com>; Cabrales, Ada <ada.cabrales@intel.com>; Lin, Shuicheng <shuicheng.lin@intel.com> Subject: Re: [Starlingx-discuss] OVS-DPDK Upgrade Testing Hello Folks, Thanks to Chris Friesen for point this out, but we believe the issues you are experiencing is due to the requirement for guests to be backed by huge pages to operate with OVS-DPDK vhost-user based ports/interfaces. The master (default) behavior for the latest nova will default to 4K pages for the guest, but this is not compatible with OVS-DPDK. The guests must be configured to use a flavor that has the property hw:mem_page_size=large set. You can follow this link to read more about the requirements on the guests for OVS-DPDK: https://docs.openstack.org/neutron/rocky/admin/config-ovs-dpdk.html “vhost-user requires file descriptor-backed shared memory. Currently, the only way to request this is by requesting large pages. This is why instances spawned on hosts with OVS-DPDK must request large pages”. Hope this helps. -Matt From: "Xu, Chenjie" <chenjie.xu@intel.com<mailto:chenjie.xu@intel.com>> Date: Tuesday, April 2, 2019 at 11:48 PM To: "Peters, Matt" <Matt.Peters@windriver.com<mailto:Matt.Peters@windriver.com>>, "Liu, ZhipengS" <zhipengs.liu@intel.com<mailto:zhipengs.liu@intel.com>>, "He, Yongli" <yongli.he@intel.com<mailto:yongli.he@intel.com>> Cc: "'starlingx-discuss@lists.starlingx.io'" <starlingx-discuss@lists.starlingx.io<mailto:starlingx-discuss@lists.starlingx.io>>, Ghada Khalil <Ghada.Khalil@windriver.com<mailto:Ghada.Khalil@windriver.com>>, "Zhao, Forrest" <forrest.zhao@intel.com<mailto:forrest.zhao@intel.com>>, Brent Rowsell <Brent.Rowsell@windriver.com<mailto:Brent.Rowsell@windriver.com>>, "Gauld, James" <James.Gauld@windriver.com<mailto:James.Gauld@windriver.com>>, "Le, Huifeng" <huifeng.le@intel.com<mailto:huifeng.le@intel.com>>, "Martinez Monroy, Elio" <elio.martinez.monroy@intel.com<mailto:elio.martinez.monroy@intel.com>>, "Perez, Ricardo O" <ricardo.o.perez@intel.com<mailto:ricardo.o.perez@intel.com>>, "Cabrales, Ada" <ada.cabrales@intel.com<mailto:ada.cabrales@intel.com>>, "Lin, Shuicheng" <shuicheng.lin@intel.com<mailto:shuicheng.lin@intel.com>> Subject: RE: [Starlingx-discuss] OVS-DPDK Upgrade Testing Hi Matt, Zhipeng is working on “PCI Affinity dependency on Nova NUMA topology”. This may relate to BUG “VM can't send packet through vhostuser port due to missing numa settings in domain xml”: https://bugs.launchpad.net/starlingx/+bug/1820378 According to him, after cutting over to nova master, the NUMA topology and PCI device info has been removed from nova. Before nova master, StarlingX uses nova stage which has NUMA topology. Some detailed information are listed below: • PCI Affinity dependency on Nova NUMA topology - Zhipeng o This affinity agent could not get both pci_device info and numa info of server from nova. o In nova stage version, we added below two for server. o server["wrs-res:topology"] o server["wrs-res:pci_devices"] o In nova master, no these attributions for server. o For topology, I can see that there is a patch of adding numa topology pending for merge o https://review.openstack.org/#/c/621476 Add server sub-resource topology API o Next step - Zhipeng to investigate alternative implementations that don't have dependencies on Nova. Best Regards, Xu, Chenjie From: Khalil, Ghada [mailto:Ghada.Khalil@windriver.com] Sent: Tuesday, April 2, 2019 11:50 PM To: Xu, Chenjie <chenjie.xu@intel.com<mailto:chenjie.xu@intel.com>>; Martinez Monroy, Elio <elio.martinez.monroy@intel.com<mailto:elio.martinez.monroy@intel.com>>; Peters, Matt <Matt.Peters@windriver.com<mailto:Matt.Peters@windriver.com>>; Lin, Shuicheng <shuicheng.lin@intel.com<mailto:shuicheng.lin@intel.com>>; Cabrales, Ada <ada.cabrales@intel.com<mailto:ada.cabrales@intel.com>>; Perez, Ricardo O <ricardo.o.perez@intel.com<mailto:ricardo.o.perez@intel.com>> Cc: 'starlingx-discuss@lists.starlingx.io' <starlingx-discuss@lists.starlingx.io<mailto:starlingx-discuss@lists.starlingx.io>>; Zhao, Forrest <forrest.zhao@intel.com<mailto:forrest.zhao@intel.com>>; Rowsell, Brent <Brent.Rowsell@windriver.com<mailto:Brent.Rowsell@windriver.com>>; Gauld, James <James.Gauld@windriver.com<mailto:James.Gauld@windriver.com>> Subject: RE: [Starlingx-discuss] OVS-DPDK Upgrade Testing Thanks Chenjie. Can you please run the following on your compute nodes and attach the output? sudo lscpu sudo virsh capabilities Richardo / Juan P, Please provide the cpu models and the above output from your two hardware systems as well. Thanks, Ghada From: Xu, Chenjie [mailto:chenjie.xu@intel.com] Sent: Tuesday, April 02, 2019 2:55 AM To: Khalil, Ghada; Martinez Monroy, Elio; Peters, Matt; Lin, Shuicheng; Cabrales, Ada; Perez, Ricardo O Cc: 'starlingx-discuss@lists.starlingx.io'; Zhao, Forrest; Rowsell, Brent Subject: RE: [Starlingx-discuss] OVS-DPDK Upgrade Testing Hi Ghada, Compute-0 and compute-1 don’t have skylake processors. The processors both are: Intel(R) Xeon(R) CPU E5-2680 0 @ 2.70GHz Compute-0 and compute-1’s server model both are: Manufacturer: Intel Corporation Product Name: S2600JF Sub-NUMA Clustering doesn’t exist in my BIOS. Only NUMA Optimized exists. An image for “Memory RAS and Performance Configuration” has been attached. Best Regards, Xu, Chenjie From: Khalil, Ghada [mailto:Ghada.Khalil@windriver.com] Sent: Tuesday, April 2, 2019 6:52 AM To: Xu, Chenjie <chenjie.xu@intel.com<mailto:chenjie.xu@intel.com>>; Martinez Monroy, Elio <elio.martinez.monroy@intel.com<mailto:elio.martinez.monroy@intel.com>>; Peters, Matt <Matt.Peters@windriver.com<mailto:Matt.Peters@windriver.com>>; Lin, Shuicheng <shuicheng.lin@intel.com<mailto:shuicheng.lin@intel.com>>; Cabrales, Ada <ada.cabrales@intel.com<mailto:ada.cabrales@intel.com>>; Perez, Ricardo O <ricardo.o.perez@intel.com<mailto:ricardo.o.perez@intel.com>> Cc: 'starlingx-discuss@lists.starlingx.io' <starlingx-discuss@lists.starlingx.io<mailto:starlingx-discuss@lists.starlingx.io>>; Zhao, Forrest <forrest.zhao@intel.com<mailto:forrest.zhao@intel.com>>; Rowsell, Brent <Brent.Rowsell@windriver.com<mailto:Brent.Rowsell@windriver.com>> Subject: RE: [Starlingx-discuss] OVS-DPDK Upgrade Testing Hi Chenjie, Can you let us know the model of your server? Does it have skylake processors? Can you also double-check that you have Sub-NUMA Clustering disabled in the BIOS? On a wolfpass, the BIOS setting is at: Enter Setup
Advanced Memory Configuration Memory RAS and Performance Configuration Sub-NUMA Cluster <Disabled>
Thanks, Ghada From: Xu, Chenjie [mailto:chenjie.xu@intel.com] Sent: Wednesday, March 27, 2019 9:59 PM To: Martinez Monroy, Elio; Peters, Matt; Khalil, Ghada; Lin, Shuicheng; Cabrales, Ada; Perez, Ricardo O Cc: 'starlingx-discuss@lists.starlingx.io'; Zhao, Forrest; Rowsell, Brent Subject: RE: [Starlingx-discuss] OVS-DPDK Upgrade Testing Hi Matt, I run the following commands on compute-0. The outputs of commands have been attached. sudo /usr/sbin/dmidecode > dmidecode.txt virsh nodeinfo > nodeinfo.txt /usr/bin/topology > topology.txt grep -i numa /var/log/dmesg > dmesg.txt And the StarlingX is installed with standard 0322 ISO image: http://mirror.starlingx.cengn.ca/mirror/starlingx/master/centos/20190322T174... Best Regards, Xu, Chenjie