Bad Network Performance for KVM Guests

[planet-itservices]

Hey Folks,

we're having problems with our network connection's to the VM's...

When trying to copy data on Shares on the VM's we receive transfer speeds with about 8 MB/s. It doesn't matter how large the files (I've tried a 1.1 GB File) are...
And it also doesn't matter if I try to copy the file to an Windows Guest or an Linux Guest.
The Connection from the Hostsystem to our Network is an 802.3ad Bonding with 4x 1Gbps Connections to an Cisco Switch where our clients are connected to.
The Problem can't be the Storage System where the Virtual HDD's from the VMs are located, an direct Data Transfer to the storage runs at full 1Gbps speed (about 90 MB/s).
We've tested the network connection between our workstations, also fine @ 90 MB/s.
But there's something strange: With NetIO from Workstation to VM's we have normal speed like from Workstation to Workstation.

At first i thought it's a windows-problem but after the check with the linux guests it seems to be a proxmox problem?
Is there any known problem and a workaround? An no, changing the network devices from virtio to rtl8139 or e1000 doesn't solve the issue.

Thanks in Advance,
Best Regards
Planet ITservices

 

[WaltervdSchee]

That is some strange thing.
On my simple Core i5 based ProxMox VE server, transfer rates are at near wirespeed, 115 MB/s to and from a Windows 7 Based workstation.
I remember going from an old Linux Samba2 based config to this new Samba 3 based config.
the tuning parameter that really helped performance on Samba2 was really hurting performance on Samba3, which was the 'socket_options' parameters.
Simply removing the socket_options parameters from the SAMBA config helped performance a lot (for me, YMMV)

 

[spirit]

try some registry tunning from here:
http://www.linux-kvm.org/page/WindowsGuestDrivers/kvmnet/registry

 

[planet-itservices]

I don't understand why it could be a Samba problem... The shares are distributed from Windows Machines, not from the host or the SAN itself.
Looks like a problem from the virtio-subsystem because its the same behalf on windows as on linux machines.

 

[planet-itservices]

Hey Guys,

I've tried an NetIO Benchmark from my Machine to one of our virtualized server and this is what I get:

UDP connection established.
Packet size  1k bytes:  24.79 MByte/s (73%) Tx,  58.18 MByte/s (0%) Rx.
Packet size  2k bytes:  7122.49 KByte/s (62%) Tx,  6448.25 KByte/s (0%) Rx.
Packet size  4k bytes:  15.43 MByte/s (49%) Tx,  11.00 MByte/s (0%) Rx.
Packet size  8k bytes:  24.16 MByte/s (45%) Tx,  19.63 MByte/s (0%) Rx.
Packet size 16k bytes:  33.06 MByte/s (49%) Tx,  32.24 MByte/s (0%) Rx.
Packet size 32k bytes:  7093.12 KByte/s (79%) Tx,  43.08 MByte/s (0%) Rx.
Done.

TCP connection established.
Packet size  1k bytes:  68.26 MByte/s Tx,  60.24 MByte/s Rx.
Packet size  2k bytes:  73.36 MByte/s Tx,  50.57 MByte/s Rx.
Packet size  4k bytes:  60.60 MByte/s Tx,  65.37 MByte/s Rx.
Packet size  8k bytes:  62.80 MByte/s Tx,  93.41 MByte/s Rx.
Packet size 16k bytes:  68.64 MByte/s Tx,  100.76 MByte/s Rx.
Packet size 32k bytes:  68.38 MByte/s Tx,  99.61 MByte/s Rx.
Done.

Strange Behalf imho... So it looks like the connection itself is performing not that bad, but file tranfsers are insanly slow, like 8 Megs per second?

 

[planet-itservices]

Push

Hey Guys, any idea about that problem? Its the same behalf like the last post. Slow File Transfers and RDP Connections but fast Transfers to the storage itself.

 

[udo]

Push

Hey Guys, any idea about that problem? Its the same behalf like the last post. Slow File Transfers and RDP Connections but fast Transfers to the storage itself.

Hi,
does the same happens without bond?

Udo

 

[blurrrr]

Push

Hey Guys, any idea about that problem? Its the same behalf like the last post. Slow File Transfers and RDP Connections but fast Transfers to the storage itself.

If there is a Firewall/Router-VM between... maybe an integrated service (like IDS) slow down the connection? Depends on the throughput of the misc. services.... Capture the packets and take a look (take a look on firewall/router-logs too)

---
EDIT: i have a dual-wan-router with 2gbit-ports (netgear)- throughput is about 40mbit cause the cpu can't handle more

 

[macday]

I also disabled flow control on all my interfaces...this helped in a mixed network 100/1000/10000 ...I just left it enabled on my drbd-interfaces