Ask Your Question
0

slow VMs on compute node [closed]

asked 2017-07-12 02:45:09 -0500

AB239 gravatar image

Hello,

I am using OpenStack Newton release with following configuration: 1 Virtual Controller node 2 Virtual Swift nodes 2 Physical Compute Nodes (Node A: 96GB RAM , 24 CPUs [Intel(R) Xeon(R) CPU E5-2420 0 @ 1.90GHz ]; Node B: 128GB RAM, 32 CPUs [Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.10GHz])

Problem Statement:

After launching a stack of around 30 instances through Horizon, VMs get distributed to Compute Nodes A and B. To my surprise, VMs on compute A complete their cloud-config in no time (approx 300 seconds) while the same cloud-config completion time in VM on compute node B takes around 2000 seconds.

Nova-compute.conf are similar on both the Compute nodes and so are nova.conf files. What could be possible cause?

Regards AB

edit retag flag offensive reopen merge delete

Closed for the following reason question is not relevant or outdated by AB239
close date 2018-03-16 02:35:03.354501

add a comment

1 answer

Sort by ยป oldest newest most voted
0

answered 2017-07-12 15:40:48 -0500

Laurent_bcn gravatar image

updated 2017-07-12 15:43:00 -0500

The virt type should be kvm in the conf file :

/etc/nova/nova-compute.conf

[DEFAULT]
compute_driver=libvirt.LibvirtDriver
[libvirt]
virt_type=kvm

And enable Intel VT and AMD-V virtualization hardware extensions in BIOS

edit flag offensive delete link more

Comments

Hi.. I do have virt_type as kvm in nova-compute on both the compute nodes.. Intel VT and AMD-V virtualization extensions are also enabled.

AB239 gravatar imageAB239 ( 2017-07-14 02:50:09 -0500 )edit

Also cat /proc/cpuinfo | grep vmx returns an output but cat /proc/cpuinfo | grep svm doesn't. Does this mean either all hardware extensions aren't enabled. But I have Intel processors on both compute nodes so don't think SVM extension will be available in BIOS.

AB239 gravatar imageAB239 ( 2017-07-14 02:53:50 -0500 )edit

Hello, In order to be sure that is a cpu issue, could you make a little bench on VMs to see if there is a difference,:

dd if=/dev/zero bs=1M count=1024 | md5sum
Laurent_bcn gravatar imageLaurent_bcn ( 2017-07-17 18:28:20 -0500 )edit

compute 1(faster): 

1024+0 records in
1024+0 records out
1073741824 bytes (1.1 GB) copied, 4.05739 s, 265 MB/s

compute 2(actually slower) : 

1024+0 records in
1024+0 records out
1073741824 bytes (1.1 GB) copied, 2.90885 s, 369 MB/s
AB239 gravatar imageAB239 ( 2017-07-19 08:23:43 -0500 )edit

So this is not a CPU issue ...

Could you check the access to the metadata url which is used by cloud-init :

   time  curl http://169.254.169.254/latest/meta-data/
Laurent_bcn gravatar imageLaurent_bcn ( 2017-07-19 09:00:36 -0500 )edit
see more comments

Get to know Ask OpenStack

Resources for moderators

Question Tools

1 follower

subscribe to rss feed

Stats

Asked: 2017-07-12 02:45:09 -0500

Seen: 531 times

Last updated: Jul 12 '17

Related questions

"Newton" - Filter RetryFilter returned 0 hosts

Cannot launch VMs with KVM since Icehouse upgrade (100% CPU) [closed]

Live migration failing with qemu error "does not support drive-mirror" [closed]

Compute nodes always showing MessagingTimeOut errors

Getting error Nova Docker container on newton

What is the difference between virt_type = kvm and virt_type =qemu in IceHouse ?

Provider Network subnet issue

Can provider network type be changed?

Nova-common not getting installed via apt-get

instance cannot be deleted from the hypervisor