Ask Your Question
0

Could you help with that odd network behavior?

asked 2013-03-25 10:49:51 -0500

Hikhvar gravatar image

updated 2013-04-10 18:05:44 -0500

smaffulli gravatar image

Hello everyone,

I use Openstack Quantum for my network management and set up an deployment accordingly to this guide: OpenStackFolsomInstallGuideWebVersion.rst from github. (Not allowed to post links)

Everything is working well except the network. I can ping the VMs over the associated floating IPs and also connect via SSH. Put the results of the ping are very odd. They start at 200ms and went down to 100ms. Each ping takes 1ms less than the previous. After reaching a ping of 100ms it goes up to 200ms again. The output of an ping can be found at the end of the question.

I pinged the Floating IP 192.168.100.105 from another VM with the floating IP 192.168.100.104. Also the data rate of an download into this VM is very slow. Although the network hardware between the web server and the VM provides at least 100Mb/s the download proceeds with 600KB/s.

Could you give me some hints, where to tweak my installation?

Regards, Hikhvar

The ping output, since I have not enough karma to post links:

ping 192.168.100.105 
PING 192.168.100.105 (192.168.100.105) 56(84) bytes of data. 
64 bytes from 192.168.100.105: icmp_req=1 ttl=63 time=168 ms
64 bytes from 192.168.100.105: icmp_req=2 ttl=63 time=155 ms
64 bytes from 192.168.100.105: icmp_req=3 ttl=63 time=154 ms 
64 bytes from 192.168.100.105: icmp_req=4 ttl=63 time=153 ms 
64 bytes from 192.168.100.105: icmp_req=5 ttl=63 time=152 ms 
64 bytes from 192.168.100.105: icmp_req=6 ttl=63 time=151 ms 
64 bytes from 192.168.100.105: icmp_req=7 ttl=63 time=150 ms 
64 bytes from 192.168.100.105: icmp_req=8 ttl=63 time=149 ms 
64 bytes from 192.168.100.105: icmp_req=9 ttl=63 time=148 ms 
64 bytes from 192.168.100.105: icmp_req=10 ttl=63 time=146 ms 
64 bytes from 192.168.100.105: icmp_req=11 ttl=63 time=145 ms 
64 bytes from 192.168.100.105: icmp_req=12 ttl=63 time=144 ms 
64 bytes from 192.168.100.105: icmp_req=13 ttl=63 time=143 ms 
64 bytes from 192.168.100.105: icmp_req=14 ttl=63 time=142 ms 
64 bytes from 192.168.100.105: icmp_req=15 ttl=63 time=141 ms 
64 bytes from 192.168.100.105: icmp_req=16 ttl=63 time=140 ms 
64 bytes from 192.168.100.105: icmp_req=17 ttl=63 time=139 ms 
64 bytes from 192.168.100.105: icmp_req=18 ttl=63 time=138 ms 
64 bytes from 192.168.100.105: icmp_req=19 ttl=63 time=137 ms 
64 bytes from 192.168.100.105: icmp_req=20 ttl=63 time=136 ms
64 bytes from 192.168.100.105: icmp_req=21 ttl=63 time=134 ms 
64 ...
(more)
edit retag flag offensive close merge delete

Comments

Please rephrase the title as a question see http://stackoverflow.com/faq#howtoask

chmouel gravatar imagechmouel ( 2013-03-26 03:45:58 -0500 )edit

1 answer

Sort by ยป oldest newest most voted
1

answered 2013-03-27 10:39:32 -0500

Salvatore gravatar image

Hi Hikhvar,

Thanks for using ask.openstack.org! I think there's more than just Quantum in understanding the dynamics you are reporting; beyond the Quantum plugin, the hypervisor, and possibly also the rest of the network infrastructure might play a role. One thing that might explain the sudden increase you are observing on ping (which goes from 100ms to 199ms in your post) is the fact that the ARP cache on the guest VM expires, and so a new ARP request is sent out.

On the slowness in bandwidth you are reporting, that is extremely hard to explain if we don't know something more about your setup. Are you using libvirt/KVM with the OVS plugin, or something else? GRE encapsulation imposes a 3% overhead; on top of this another factor impacting performance if that with GRE encapsulation, the NIC is not able to leverage TSO. It might also be good to know how you are measuring the performance.

Finally, I am sorry but I have no answer at the moment for the slowly decreasing RTT times you are observing.

edit flag offensive delete link more

Comments

Hey, we are using libvirt/KVM with the OVS plugin. Our setup is based on this github.com/mseknibilel/OpenStack-Folsom-Install-guide/blob/master/OpenStackFolsomInstallGuideWebVersion.rst . We have merged the ControlNode and the NetworkNode since we run a small testbed. Could this be the reason?

Hikhvar gravatar imageHikhvar ( 2013-04-04 05:46:51 -0500 )edit

Your Answer

Please start posting anonymously - your entry will be published after you log in or create a new account.

Add Answer

Get to know Ask OpenStack

Resources for moderators

Question Tools

Stats

Asked: 2013-03-25 10:49:51 -0500

Seen: 357 times

Last updated: Mar 27 '13