VMware Cloud Community
sketchy00
Hot Shot
Hot Shot

speed of NIC's on two VM's on the same host

It seems to be well documented that when two VM's are talking to eachother via their vNIC's and they reside on the same host and vswitch, they in fact use the system bus instead of across the ntework infrastructure (as long as there is no vlan tagging), thus making the communication much faster. This is such an intriguing concept that I've been looking into it more, but have been stumped at being able to determine how much of an improvement one would see over say, host systems with 2 of their 6 NICs dedicated for VM/LAN communication connected to traditional GigE switchgear. Does anyone have any experience with testing out the difference on this?

Thanks,

Sketchy.

Tags (2)
0 Kudos
5 Replies
Texiwill
Leadership
Leadership

Hello,

There may be a performance difference and there may not be, it all depends on the load on the host in question. Network loads that communicate through the same portgroup on the same vSwitch (or 2 portgroups on the same vSwitch that have the same VLAN ID) do not go through the pNICs but stay within the vSwitch, however, this may not imply you get more than 1 Gb of transfer speed. One way to test this would be to use something like netperf. But I would not do this unless they were the only 2 VMs on the host. Then do the same test but target an external system through your switching network.

The key is that these measurements will give you the MAXIMUM VM to VM and VM to Physical you can achieve but once you load a system with more than these VMs overcommit comes into play and you have scheduling and switching coming into play that may reduce overall 'expected' network performance.


Best regards,

Edward L. Haletky VMware Communities User Moderator, VMware vExpert 2009, Virtualization Practice Analyst[/url]
Now Available: 'VMware vSphere(TM) and Virtual Infrastructure Security: Securing the Virtual Environment'[/url]
Also available 'VMWare ESX Server in the Enterprise'[/url]
[url=http://www.astroarch.com/wiki/index.php/Blog_Roll]SearchVMware Pro[/url]|Blue Gears[/url]|Top Virtualization Security Links[/url]|Virtualization Security Round Table Podcast[/url]

--
Edward L. Haletky
vExpert XIV: 2009-2023,
VMTN Community Moderator
vSphere Upgrade Saga: https://www.astroarch.com/blogs
GitHub Repo: https://github.com/Texiwill
0 Kudos
sketchy00
Hot Shot
Hot Shot

Thanks Ed,

I suppose testing is truly the only way to see if there is a real-world difference. I was hoping that there might be some anecdotal evidence that might suggest a noticable improvement. My use-case is this. I will have probably 20 or so VM's spread across 4 ESX hosts (or relegated to 1 ESX host if it behooves us to do so) that peform code compling for us. The reason for so many machines is because that is about the number of platforms and bit levels we support. They are mostly linux VM's, and will have their home directories (where the compiled bits are dumped out) located on another VM that will serve those locations up via NFS. This somewhat mirror's the physical environment that those build machines currently run on. If the VM serving up the NFS mount points is contending too much, I may be able to use Equallogic's guest ISCSI initiator to take advantage of some MPIO, but will keep it simple until I see there is a bottle-neck.

Is it a correct statement that when traffic is indeed being sent across the system bus, it may be faster than what a 1gb traditional wired connection can achieve? If I can demonstrate even some minimal improvement, that always helps the "buy-in" effect of the always suspicious development crew.

0 Kudos
LucasAlbers
Expert
Expert

I put all of my code tree's on a physical nfs/samba server, as I find that I saturate the disk io.

0 Kudos
Texiwill
Leadership
Leadership

Hello,

Is it a correct statement that when traffic is indeed being sent across the system bus, it may be faster than what a 1gb traditional wired connection can achieve? If I can demonstrate even some minimal improvement, that always helps the "buy-in" effect of the always suspicious development crew.

I would run the tests to see if it is possible.


Best regards,

Edward L. Haletky VMware Communities User Moderator, VMware vExpert 2009, Virtualization Practice Analyst[/url]
Now Available: 'VMware vSphere(TM) and Virtual Infrastructure Security: Securing the Virtual Environment'[/url]
Also available 'VMWare ESX Server in the Enterprise'[/url]
[url=http://www.astroarch.com/wiki/index.php/Blog_Roll]SearchVMware Pro[/url]|Blue Gears[/url]|Top Virtualization Security Links[/url]|Virtualization Security Round Table Podcast[/url]

--
Edward L. Haletky
vExpert XIV: 2009-2023,
VMTN Community Moderator
vSphere Upgrade Saga: https://www.astroarch.com/blogs
GitHub Repo: https://github.com/Texiwill
0 Kudos
sketchy00
Hot Shot
Hot Shot

Okay... will do.

Thanks for your input. If you are every looking for putting out another good article on all things VMWare, this would definately be a subject that would be worth doing. Just doesn't seem to have quite the swatch of information behind it as other features. ...but it's so darn intriguing.

0 Kudos