Packet Loss with KVM guests

H

hverbeek

Member
Feb 14, 2011
40
1
8
I'm running PVE 1.7 on the 2.6.32-4-pve Kernel. I have only KVM guests, running debian 5 lenny 64bit. The guest VMs use e1000 network drivers. Network setup is typical for Hetzner-hosted machines (routing from KVM Guest to the internet via PVE Host).

Problem: While investigating a network problem I ran several instances of 'mtr' on on of the KVM guests, pinging hops throughout the internet. When one 'mtr' is running, everything is fine. As soon as I start multiple instances, I see massive packet loss (in the area of 45 - 60%) at the first hop, the PVE Host.

Has anyone noticed this before?
is there a recommendation?
 
I'm confused. If you are running KVM guest, then the host should not show up in the trace at all. I see the same thing, but for me, the first hop is my gateway, not the HN. I am wondering if this is some kind of rate limiting?
 
dswartz said:
If you are running KVM guest, then the host should not show up in the trace at all.
Click to expand...

As I said, it's a Hetzner-setup where the guest subnet is routed to the KVM host (Hetzner = german hosting provider). In other words the host is the gateway.
 
Sorry, I didn't notice the routing comment in your OP. No idea, sorry. Except I see this myself, so I don't think it is unique to hetzner?
 
hverbeek said:
I'm running PVE 1.7 on the 2.6.32-4-pve Kernel. I have only KVM guests, running debian 5 lenny 64bit. The guest VMs use e1000 network drivers. Network setup is typical for Hetzner-hosted machines (routing from KVM Guest to the internet via PVE Host).

Problem: While investigating a network problem I ran several instances of 'mtr' on on of the KVM guests, pinging hops throughout the internet. When one 'mtr' is running, everything is fine. As soon as I start multiple instances, I see massive packet loss (in the area of 45 - 60%) at the first hop, the PVE Host.

Has anyone noticed this before?
is there a recommendation?
Click to expand...

Is there any reason to NOT USE virtio ?


michu
 
Hi,

I really don't have idea what happens. I have 2 proxmox servers with most ubuntu 10.04 server quests and few windows xp/7, all KVM and no problems with Linux on virtio and Windows on e1000.

Maybe it's hardware/driver problem ?

michu
 
Which kernel are you running on the Proxmox host? I'm (still) on 2.6.32-4, wondering if I should upgrade...
 
hello,

unfortunately we have the same problem here.
is there any known solution?

edit:
kernel: Linux Proxmox-VE 2.6.35-1-pve
guests: 2.6.32-5-amd64 (debian)

testet different (virtual) ethernet hardware, already - no significant difference.

thanks in advance.
best, achim.
 
Last edited:
move to our latest 2.6.32.
 
hello tom,
thanks for your reply.

do you know if it is safe to switch to proxmox-ve-2.6.32?

best,
achim.
 
I do not understand your question. 2.6.32 is the recommended kernel as already written.
 
Hey Tom,

sorry. Actually "Linux Proxmox-VE 2.6.35-1-pve" is installed on my system.
It seems to be the default in "Hetzner's" Proxmox image.

Of course I can easily install 2.6.32 but I'm not sure if this will break my system.
This is why I asked "if it's safe" ;)

Thanks in advance!

Best,
Achim
 
if you run a hosted system you should ask Hetzner if their hardware support our kernels - how should we know this?

btw, Hetzner is no Proxmox partner so there is no knowledge here about their Proxmox VE use. I highly recommend to go for a Proxmox Hosting partner - they have deep knowledge and experience with Proxmox VE and also direct contact to our development and also quite important, they also support the project with money.

see http://www.proxmox.com/partners/hosting-providers/featured-partners
 
I'm running Proxmox VE on Hetzner EQ10 for a while now. Our only difference from their stock hardware are additional Intel GBit NICs for the cluster interconnect.

In my eyes, 2.6.32 is a kernel version deemed for long-term-support. Since the hardware is running fine, and since I don't need any other features (e.g. KSM), I see no need for a different kernel version.

The latency is solved for us. We use virtio network devices for everything. Almost all guests are Debian 6 squeeze, for the few Debian 5 lennys we used the 2.6.32-bpo kernel. I find that Squeeze guests perform much better than Lenny guests.
 
our latest 2.6.32-6 has all needed features, also KSM and latest intel drivers (e1000e, igb, ...)
 
You must log in or register to reply here.
Top Bottom