New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Any Proxmox pros here?
I have performance issue with my VM's network speed. Could use some guidance in fixing this. =/
My issue is as it follows: Host machine gets 800/600 Mbps, while any VM's are only getting 5-20 Mbps. I can't figure out where the issue lies. If you someone could take a look, it would help me alot.
Comments
Try using a different network adapter (e.g. E1000).
probably much more infos needed esp. about what configuration (nat, routed, bridged) is used and what virtualization and OS the guests are on...
I am not really considering me an expert, more likely intermediate, but willing to try and help, so some first questions:
do you use other then virtio drivers?
how do you measure those speeds and could any set I/O-limit cause problems?
have you tried setting up a default lxc-guest without any fancy config settings just in case?
Host machine (OVH Proxmox 4.3 default install):
KVM VM (VirtIO) (Ubuntu 16.04 with no extra settings) Bridged with OVH Failover IP
There are no I/O issues or limits because in the above scenario nothing is running on the host and there are no other VM's even created.
Try with E1000
Not any better..
that's weird. with bridged setup you probably set up a virtual mac with OVH and all that?
I'd probably try to add the IP on the hostnode and use a routed setup just for testing purposes or connect the guest via nat to see speeds then.
may be the IP itself is somewhat bad routed by OVH? what did you use as gateway and such?
What bridging are you using, the Linux bridging or OpenVswitch?
also is the firewall enabled?
Fine here. Can reach about 30 MB/s
that's weird. with bridged setup you probably set up a virtual mac with OVH and all that?
Yes. I created virtual mac for failover ip's and used that on the VM settings
I'd probably try to add the IP on the hostnode and use a routed setup just for testing purposes or connect the guest via nat to see speeds then.
I tried 1 IP setup (NAT) and the speeds are perfect. As soon as I use failover IP for the VM's (CT or KVM), the speeds are horrible.
what did you use as gateway and such?
My host machine IP is 137.74.1.xx so I used 137.74.1.254 as gateway as it is specified in OVH proxmox guide.
You have WAW Discovery server with WAW Failover IPs? If so, tell me exactly how did you set it up to get those speeds for your VM's? Because the way I have it right now (Default install) worked perectly on SoYouStart server.
--2016-12-03 19:43:00-- http://ovh.net/files/1Gio.dat
Resolving ovh.net (ovh.net)... 213.186.33.6
Connecting to ovh.net (ovh.net)|213.186.33.6|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 1073741824 (1.0G) [application/octet-stream]
Saving to: ‘1Gio.dat’
Centos 7 box behind pfSense, Nat, onto Vrack on one of the Xeon-D's
Using OpenVswitch and Virt IO, Proxmox 4.3
Sorry but what is WAW?
I just use Proxmox 4 template from OVH. They use bridge setting. I add virtual mac and use VirtIO for disk and network.
But when I use sata0 for disk and virtio for network, I can get 30 MB/s on the first 5 second, and after that I just get 2 MB/s. I think this is issue in IO to disk caused by the use of sata0.
This is not the new Discovery server in OVH Poland Datacenter at Warsaw.
I did it to.. In SoYouStart and it worked just fine.. However on the Warsaw server it does not. Maybe they have misconfigured the failover ips? I mean it just went live 2 days ago.
I see. Sorry I don't have server on WAW.
Maybe you can try another virtualization from ovh, and then test again, or try to add one additional IP
I did. Hyper-V using my school's lisence. Proxmox 3.4, ESXi 6.0. All of them have this same issue. I'm starting to think that OVH network is fucked up in their new datacenter.
How about another additional IP? It worth to try
Ive tested them all lol.. Some in LXC conatiner and some in KVM machines.
All of them have similary low speeds. Which is why I'm looking for others who have WAW server and WAW failovers to test it.
There must be an option to increase the bitrate for the container.
......
I have them unlimited by default
have you done some traceroutes to the testserver from the host vs the guest? Any peculiarities when compared?
After reading it's on their new DC I'd guess the files you wget are simply pulled from different servers. Their internal cdn might choose the nearest box based on whatever db-lookup for the IP. so if your addon IP gets you a testfile in france rather then poland that's pretty much it.
I did speedtest-cli multiple times and it had same speeds as well. I tested multiple locations.
Just to be entirely sure and for the future, you could try to run
wget -O /dev/null <url>
which downloads the file into /dev/null (discard-device) instead of writing it to the disk consuming disk I/O.I had something odd at one point with OVH. If you mtr do you get packet loss to and from the vm? If you do restart the networking process and try again if it is normal until about 20 mins are up and drops down then reinstall the hostnode. It was a strange issue I had and I gave up troubleshooting and just reinstalled and it was perfect afterwards.
You've probably tried the above anyway but just throwing the idea out there.
--snip just ignore this, didn't use my brain first.
Its wayne or? u install the proxmox Kernel anyway
I still believe you are getting testfiles from different cdn servers assigned, depending on the IP used...
Please do a more complete test like https://github.com/K4Y5/ServerBench from host and guest and compare the results...
I did iperf between host and the KVM and this is the result:
The connection from KVM to host
On host: iperf -s
On KVM: iperf -c 137.74.xx.xx -i 1
Looks all good from KVM to Host (Public IP used in test)
Next I test the connection from host to KVM:
On KVM: iperf -s
On host: iperf -c 193.70.95.xx -i 1
Again I used public IP's. Now we can conclude that there is no performance loss when using iperf it seems.
Now I run the serverbench you linked.
On KVM:
On host:
From here we can see that there is a huge difference in EU speeds and that the KVM does not get fast speeds inside EU like the host does. The KVM has caching enabled so the IO is faster.
Theres not much difference in Asia/USA speeds since peering towards there is shit anyway. It's the EU speeds im worried about. The speed loss is over 90 MB/s for some EU locations on KVM.
That is the different routing of the additional IPs - Traceroute/MTR should show it.
I've got the same problem on VMs, changed system to ESXI6.0 problem solved, no clue why
What settings did you use on ESXI 6.0?