Poor network performance

Discussions related to using VirtualBox on Solaris hosts.
Post Reply
mattjk
Posts: 2
Joined: 9. Sep 2009, 14:50
Primary OS: OpenSolaris 11
VBox Version: PUEL
Guest OSses: W2K8, W7

Poor network performance

Post by mattjk »

Hi all,

I'm looking for some help diagnosing poor network performance in VirtualBox.


Firstly, my environment:

Intel-branded workstation board with onboard Intel PRO/1000 NIC
Core 2 Duo
8GB RAM
Running OpenSolaris 2009.06 and VirtualBox 3.1.8
Gigabit networking

VirtualBox is hosting four VMs - 3 x Windows 2008 (non-R2) x64, 1 x Windows 7 x64. All are single vCPU, have 1-2GB of RAM, and a single vNIC bridged to the PRO/1000 pNIC.


I am experiencing poor network performance copying files to/from the guests via SMB - best case scenario is about 5MB/s average, typically closer to 3MB/s average. The problem affects all guests equally, and occurs any time a VBox guest is involved in either end of the transfer (e.g. guest -> guest, guest -> physical, physical -> guest all cap out at 3-5MB/s).

I've done as much troubleshooting as I know how and have gotten nowhere - I can say what it's not though:

- physical network problems (physical -> physical file copies are MUCH faster)
- network congestion - my environment is very light load (in every respect)
- disk I/O performance problems - file copies INSIDE the VMs run at 20+ MB/s
- CPU-related - CPU usage is low during network copies

Ideas anyone?

Thanks,
Matt
kebabbert
Volunteer
Posts: 321
Joined: 31. May 2008, 10:00
Primary OS: OpenSolaris 11
VBox Version: OSE other
Guest OSses: WinXP, RedHat, Ubuntu

Re: Poor network performance

Post by kebabbert »

No idea. But you should upgrade OpenSolaris to later builds, and also try CIFS instead. CIFS is faster than SMB.
hvr_za
Posts: 10
Joined: 21. Jul 2010, 21:36
Primary OS: OpenSolaris 11
VBox Version: OSE other
Guest OSses: WIndows 2008 R2 windows SBS 2008

Re: Poor network performance

Post by hvr_za »

I have the same problem, It start off as 2MB/s and then increase to about 16MB/s (between guests on virtualbox). It takes about 1 min copy copy to reach 16MB/s. I have 5x 250GB SATAII 7200rpm disk in a raidz with a perc 6i controller. Hardly any disk activity during the copy. Somehow 16MB/s does not seem right when each VM is connected at 1Gbps. Both nodes are bridged.
When I ping other devices on the network it sometimes spike up to 19ms from a guest.

DELL T610 2x 2.93GHz quad core Xeon.
16GB memory DDR3 1066
bnx nic
Opensolaris build 134
vbox 3.2.6

Can anyone please comment if they have a bridge connection without any performance issue running in Opensolaris.
nxnlvz
Posts: 28
Joined: 16. Dec 2008, 07:45
Primary OS: Solaris
VBox Version: PUEL
Guest OSses: Widnows (XP,7,8) / Linux (Debian, Unbuntu) / MacOS (Lion)

Re: Poor network performance

Post by nxnlvz »

Which network interface type are you using for the Guest? What does the Guest tell you that the interface speed is?

-nxn
bqbauer
Posts: 159
Joined: 14. Dec 2008, 22:04

Re: Poor network performance

Post by bqbauer »

hvr_za wrote:Can anyone please comment if they have a bridge connection without any performance issue running in Opensolaris.
First, make sure you're not mixing up KB/s and Kb/s. With networking overhead, the conversion is closer to 10x than 8x. So 3MB/s is about 30Mb/s, which is how network speeds are generally measured. Still, that's slow.

I have no performance issues with bridged or NAT networking on my OpenSolaris host and its guests--not since 2009.06 was released with Crossbow. I don't do anything special with Crossbow, but its existence greatly improved previous performance issues with VB guests. However, I NEVER use SMB for transfers. I use either the shared folders method (which you may be calling SMB?), SSH, or FTP with my host.

I wonder if you're hitting host disk performance issues since your transfers may be between guest & host on the same physical drive? On Windows guests, try something like this, which sends it to the null device (like /dev/null on UNIX), and note the speeds I got. I got about 300Mb/s (and by the way, this is pushing the upper limit of FTP's speed throughput):

ftp vb_host
ftp> bin
ftp> get ubuntu-8.10-desktop-amd64.iso nul: (That's the word NUL with a colon)
200 PORT command successful.
150 Opening BINARY mode data connection for ubuntu-8.10-desktop-amd64.iso (732989440 bytes).
226 Transfer complete.
ftp: 732989440 bytes received in 21.24Seconds 34509.86Kbytes/sec.
ftp>

When I don't transfer to the null device I get about half:

ftp> get ubuntu-8.10-desktop-amd64.iso
200 PORT command successful.
150 Opening BINARY mode data connection for ubuntu-8.10-desktop-amd64.iso (732989440 bytes).
226 Transfer complete.
ftp: 732989440 bytes received in 42.62Seconds 17197.85Kbytes/sec.
ftp>
bqbauer
Posts: 159
Joined: 14. Dec 2008, 22:04

Re: Poor network performance

Post by bqbauer »

I just tried my example with VB shared folders by copying the same large file from my mounted drive letter (served up by the VB host) to NUL:. It took three seconds. When I copied it to my guest's "drive", it took 30 seconds. Faster than FTP when actually writing out a file locally. I feel this is fairly stellar performance considering I'm copying from the host into a virtual drive on the same disk.

I don't have an SMB server running on my host, so can't test that for you.

* My host for the above posts is an Intel Q6600 CPU at 2.7GHz, 8GB memory, and guests tested have 2GB memory, single CPU, and use the virtual Intel Pro/1000 network card. The virtual HD controller is the default IDE/PIIX4. VT-x is always enabled, and the Q6600 doesn't have nested paging.
Last edited by bqbauer on 28. Jul 2010, 07:12, edited 1 time in total.
Ramshankar
Oracle Corporation
Posts: 793
Joined: 7. Jan 2008, 16:17

Re: Poor network performance

Post by Ramshankar »

Hmm.. I wonder if this could be due to GRO packet fragmentation causing retransmissions. Which card are you using on the host (and driver name)?
Oracle Corp.
bqbauer
Posts: 159
Joined: 14. Dec 2008, 22:04

Re: Poor network performance

Post by bqbauer »

Ramshankar wrote:Hmm.. I wonder if this could be due to GRO packet fragmentation causing retransmissions. Which card are you using on the host (and driver name)?
For clarity, I edited my post just above for comparison--even though I'm sure this comment was directed at the original author of the thread.
Post Reply