VMware Cloud Community
faruge
Contributor
Contributor

Distributed Switching

Hi there,

Is anyone able to provide information regarding best practices with distributed switching when host configurations are different (i.e. NIC counts are different).

If there is an environment with say 10 hosts, each of those hosts have 8 1GB NICs - Setting up distributed switching is straight forward - Perhaps 4 vDS is created.

However, if you decide to add say 4 new hosts to the cluster, and these hosts come with 2 x 10GB NIC's - Does distributed switching become more complex?

Thanks in advance for your help Smiley Happy

Kind Regards,

Justin

10 Replies
saikumarv
Contributor
Contributor

Hi,

In general  as per networking fundamentals, it is best to configure the network cards of same speed at vm and switch level,

irrespective of heterogenous environment machines.  This is primarily to avoid performance issues / packet level communication;

DV switch configuration should not add any complex to such setup, if we are ok to compromise on performance.

Reply
0 Kudos
faruge
Contributor
Contributor

Thank you for your information - That does make perfect sense.

Just to further elaborate -

Below is the current DV switch configuration (4 in total).  Each host has 8 adapters.  Each DV Switch has 2 uplinks assigned to 2 physical adapters on each host.

pastedImage_0.png

With 4 news hosts coming in, each with only 2 network adapters, I believe assigning them to the current DV switch configuration is not going to work?  I assume I can't assign physical adapters to multiple DV Switches?  Under standard switching, I would just create 1 virtual switch and assign both adapters. Then create the necessary port groups.  This would be ok, but I know I lose management capabilities of the network unless I use DV switching.

Will I need to restructure my DV switch configuration into 1 single switch with the necessary port groups to now cater for the new hosts?

Thanks again for your help Smiley Happy

Kind Regards,

Justin

Reply
0 Kudos
saikumarv
Contributor
Contributor

Hi,

Hope you are asking to cater new hosts under a new standard switch and use them.

In my knowledge, Yes,  that would be better, but with a compromise on management of this new standard switch.

Still I asked one my of senior also to think over this, hope to get response.

daphnissov
Immortal
Immortal

but I know I lose management capabilities of the network unless I use DV switching.

What capabilities are you thinking you'll lose by putting your management vmkernel port on a vSS?

If you add more hosts with 2 x 10 GbE adapters, you should create a vDS especially for them. I wouldn't recommend trying to squeeze those in to your other vDSs.

Reply
0 Kudos
faruge
Contributor
Contributor

Hi,

The capabilities I was referring to was vDS as a whole such as Network IO Control.

From what you are saying perhaps I can create 2 vDS :

1. vDS for the servers running 8 x 1GB Networking

          a. Management Port Group

          b. vMotion Port Group

          c. iSCSI Port Group

          d. Virtual Machine Port Group

         

2. vDS for the servers running 2 x 10GB Networking

          a. Management Port Group

          b. vMotion Port Group

          c. iSCSI Port Group

          d. Virtual Machine Port Group

The only question I would have for this setup is will I be able to migrate virtual machines from a host using the 1GB vDS' to a host using the 10GB vDS' without having to change the virtual machines network adapter details as the port group names will be different?  Also, would I get proper HA in the event of hardware failures?  Will VM's automatically power on a host with a different vDS attached?

Thanks again for your assistance Smiley Happy

Kind Regards,

Justin

Reply
0 Kudos
daphnissov
Immortal
Immortal

The only question I would have for this setup is will I be able to migrate virtual machines from a host using the 1GB vDS' to a host using the 10GB vDS' without having to change the virtual machines network adapter details as the port group names will be different?  Also, would I get proper HA in the event of hardware failures?  Will VM's automatically power on a host with a different vDS attached?

I honestly have never tried this configuration and so I don't know. Even if the names of the port groups are the same, the IDs would be different. This is one reason why hosts with different hardware configurations should be kept, as much as possible, to their own cluster.

saikumarv
Contributor
Contributor

Hi,

for any vmotion, source / destination configuration would require same configuration;

I do not think with given setup will work for HA and fault tolerance.

Reply
0 Kudos
faruge
Contributor
Contributor

Ok, thats good to know.

It appears every white paper I read regarding distributed switches refers to the hardware configuration being exactly the same with regards to the NIC's.

Thanks for your help Smiley Happy

Kind Regards,

Justin

Reply
0 Kudos
jchilton
Enthusiast
Enthusiast

Justin, out of interest did you ever consider creating just 1 vDS and having multiple Uplink ports (vmnics) and segregating traffics types by vdPort Groups?

From my limited experience of vDS I have always created a vStandard Switch for vmkernel Mngt traffic and then created a single vDS with all other traffic types for NIOC control )shares/res/limits).

I do think the vDS is suited to esxi hardware that is identiical e.g no of vmnics/uplinks.  There is probably a way of making this work like adding a 10gb card in the old servers and creating a new vDS so you can transfer stuff. Good luck.

Reply
0 Kudos
faruge
Contributor
Contributor

Hey there,

I tried in a lab environment with just a single vDS but I still had issues distributing the uplinks to all the correct ports - iSCSI in particular.

I spoke to VMware support.  Their suggestion is to create a separate cluster of servers with the 10GB NICs.  It can become 'messy' if you try to mix the environments under 1 cluster.

But thanks for your suggestion - One thing I didn't consider was to just add 10GB NIC's to the existing servers.  I'll find out if this is feasible from a budgetary point of view Smiley Happy

Thanks again.

Justin

Reply
0 Kudos