Hi

We have been trying to bring up a setup with devstack+opendaylight+vpp 
components for last few weeks without much success. Sorry for putting long 
e-mail here. Hope I could convey the problem that we are facing. Could anyone 
point us if there is any fundamental issue in our setup? Appreciate any inputs 
here and plz let us know if u need any specific logs.



1.       For this setup, we are following the README instructions from the 
following devstack-vpp gerrit patch: 
https://git.opendaylight.org/gerrit/#/c/47512/14/demos/vpp-gbp-devstack/README.rst

2.       We tried to bring up these components in different setups:

a.       Setup 1:

                                                               i.      8 core, 
24GB RAM, Ubuntu 16.04

                                                             ii.      Used 
VirtualBox version 5.1 with Vagrant version 1.8.6

                                                           iii.      Used 
Vagrant file (from above gerrit patch), brought up two CentOS 7.2 VMs: One for 
devstack-controller (7GB RAM, 2CPU) and one for devstack-compute-1(7GB RAM, 
2CPU)

                                                           iv.      Installed 
VPP & HC 17.01 on both controller and compute nodes

                                                             v.      QEMU 
version on the nodes was 2.6.0

                                                           vi.      Ran ODL 
with vpp renderer inside a 3rd VM (Ubuntu)

b.       Setup 2:

                                                               i.      Very 
similar to setup1, but on higher configuration servers and VMs: 40 core, 160GB 
RAM, Ubuntu 16.04

                                                             ii.      Used 
VirtualBox version 5.1 with Vagrant version 1.8.6

                                                           iii.      Used 
Vagrant file (from above gerrit patch), brought up two CentOS 7.2 VMs: One for 
devstack-controller (16GB RAM, 8CPU) and one for devstack-compute-1(8GB RAM, 
8CPU)

                                                           iv.      Installed 
VPP & HC 17.01 on both controller and compute nodes

                                                             v.      QEMU 
version on the nodes was 2.6.0

                                                           vi.      Ran ODL 
with vpp renderer inside a 3rd VM (Ubuntu)

c.       Setup 3:

                                                               i.      Very 
similar to serup 2, except Instead of using virtualized environment for 
devstack, installed devstack directly on the physical server.

                                                             ii.      Installed 
devstack-all-in-one configuration on the physical server with Ubuntu 16.04

                                                           iii.      Installed 
VPP & HC 17.01 on the physical server

                                                           iv.      QEMU 
version on the physical server was 2.5.0

                                                             v.      Huge pages 
were bumped from 512 to 8192

3.       As specified in the README, we created a cerros image and tried it 
with different nova flavors (512MB, 1024MB, 2048MB RAM setting)

4.       When we boot up nova instances, we have noticed different types of 
nova errors at different times in different setups. Some sample error logs from 
nova daemons: In all these setups, except in setup1, there is abundant amount 
of free RAM available on the VPP nodes.

a.       On Setup1

                       i.   512MB based flavor: ^[[01;31m2017-02-02 
18:32:34.232 TRACE nova.compute.manager ^[[01;35m[instance: 
343dd3bc-d4f6-471b-ba31-57460ebd79da] ^[[00mlibvirtError: Requested operation 
is not valid: cpu affinity is not supported

                                                             ii.      1024MB 
based flavor: 
path=/dev/hugepages/libvirt/qemu,share=yes,size=1073741824,host-nodes=0,policy=bind:
 unable to map backing store for guest RAM: Cannot allocate memory

b.       On Setup2

                       i.   512MB based flavor: ^[[01;31m2017-02-13 
19:42:06.184 TRACE nova.compute.manager ^[[01;35m[instance: 
df856ed7-5cdc-48b9-8392-9fd2dbeba353] ^[[00mlibvirtError: Requested operation 
is not valid: cpu affinity is not supported

                      ii.   2048MB based flavor: ^[[01;31m2017-02-13 
18:44:20.321 TRACE nova.compute.manager ^[[01;35m[instance: 
86a38404-7abb-49ea-a6df-7696f16f479c] ^[[00m2017-02-13T18:44:19.952716Z 
qemu-kvm: -object 
memory-backend-file,id=ram-node0,prealloc=yes,mem-path=/dev/hugepages/libvirt/qemu,share=yes,size=2147483648,host-nodes=0,policy=bind:
 unable to map backing store for guest RAM: Cannot allocate memory

                    iii.   3072MB based flavor: 2017-02-13 19:55:58.310 
^[[01;35m[instance: b8fe63cb-72ef-410e-aad0-ac5cba955fb8] ^[[01;31mError from 
last host: devstack-compute-1 (node devstack-compute-1): [u'Traceback (most 
recent call last):\n', u'  File "/opt/stack/nova/nova/compute/manager.py", line 
1783,: Build of instance b8fe63cb-72ef-410e-aad0-ac5cba955fb8 was re-scheduled: 
Insufficient compute resources: Requested instance NUMA topology cannot fit the 
given host NUMA topology.\n']^[[00m

c.       On Setup3 (with huge pages bumped to 8192)

                       i.   2048MB based flavor: 2017-02-14 15:17:07.270 
^[[01;35m[instance: 6653f474-2077-4422-9582-bc54d118b86a] ^[[01;31mError from 
last host: node.sri-pod10. (node node.sri-pod10.xos-pg0.wisc.cloudlab.us): 
[u'Traceback (most recent call last):\n', u'  File 
"/opt/stack/nova/nova/compute/manager.py", line 1783,: Build of instance 
6653f474-2077-4422-9582-bc54d118b86a was re-scheduled: Insufficient compute 
resources: Requested instance NUMA topology cannot fit the given host NUMA 
topology.\n']^[[00m

Thanks
Srikanth
_______________________________________________
vpp-dev mailing list
[email protected]
https://lists.fd.io/mailman/listinfo/vpp-dev

Reply via email to