VMware Cloud Community
aaron_champion
Contributor
Contributor
Jump to solution

Can I use VDS with LACP to reduce iSCSI connections on SAN group?

I'm trying to reduce the number of iSCSI connections to my EqualLogic storage group. I am currently using a standard vSwitch with two vmk ports per server for SAN. Can I use a single vmk with VDS/LACP to reduce my iSCSI connections by half per server and still maintain bandwidth/redundancy?

I'm fairly confident this will work, but was hoping for validation. Please correct me if I'm totally off or if this would be unsupported.

Reply
0 Kudos
1 Solution

Accepted Solutions
chriswahl
Virtuoso
Virtuoso
Jump to solution

My volumes span multiple physical devices (20 paths for 7 volumes per iSCSI adapter). Can I assume that with load balancing on a LAG set to hash both source and destination, I would get at least some load balancing between adapters?

If you have multiple L2-4 target endpoints, yes. Such as an IP address, MAC address, or unique ports. Paths, volumes, and devices are not relevant to LACP.

If I do choose a LAG, I'm having trouble visualizing how it might reduce my iSCSI connections to my storage pool. I'm configuring the targets on my Broadcom adapters, the vmk connects through those adapters. Is the logic smart enough to only connect active paths on one adapter or another in a LAG? I'm not even sure I'm making sense anymore Smiley Wink I might have to play with a LAG on an isolated host for a bit and see how exactly it will behave.

The adapters are no longer exposed. The LAG forms one logical interface. Traffic can be sent or received across any physical adapter in the LAG.

Note: I would not use a LAG for iSCSI traffic on a vSphere host: Seriously, Stop Using Port Channels for vSphere Storage Traffic | Wahl Network

VCDX #104 (DCV, NV) ஃ WahlNetwork.com ஃ @ChrisWahl ஃ Author, Networking for VMware Administrators

View solution in original post

Reply
0 Kudos
6 Replies
chriswahl
Virtuoso
Virtuoso
Jump to solution

I think what you're asking is if using LACP will utilize multiple network adapters with a single VMkernel port. And the answer there is: no.

Will traffic flow to your storage array? Yes.

The ESXi host will inspect its routing table, search for a VMkernel port on the same subnet as the iSCSI target (or otherwise use the default gateway), and then throw the session over to LACP to hash out an uplink choice. This uplink choice will not change unless there is a link failure. At that point, traffic will be shifted to the surviving link.

I go over iSCSI in great detail in the Networking for VMware Administrators book.

VCDX #104 (DCV, NV) ஃ WahlNetwork.com ஃ @ChrisWahl ஃ Author, Networking for VMware Administrators
HeathReynolds
Enthusiast
Enthusiast
Jump to solution

This would maintain redundancy assuming the upstream physical switch supports multi chassis etherchannel (like Cisco VPC), but it wouldn't maintain the bandwidth of two iSCSI connections. LACP is only going to hash the connection to a single link, to take advantage of the multiple links in an LACP etherchannel you are going to need multiple IP addresses.

I haven't tried ISCSI on LACP, I know there are some checks when you bind the ISCSI vmkernel to the virtual storage adapter. I believe one of the checks is that the VMK is only active on a single link, but I'm not sure. It's been a little while since I set up ISCSI.

My sometimes relevant blog on data center networking and virtualization : http://www.heathreynolds.com
Reply
0 Kudos
chriswahl
Virtuoso
Virtuoso
Jump to solution

I haven't tried ISCSI on LACP, I know there are some checks when you bind the ISCSI vmkernel to the virtual storage adapter. I believe one of the checks is that the VMK is only active on a single link, but I'm not sure. It's been a little while since I set up ISCSI.

The checks verify that only a single network adapter is marked Active and that all other adapters are marked Unused. Keep in mind, however, that Enhanced LACP with a 5.5 VDS creates a single uplink object to represent the LAG - it no longer uses the individual network adapters in the teaming policy.

It also verifies that the storage array is layer 2 adjacent by checking the hops to the target (routed iSCSI binding is not supported).

VCDX #104 (DCV, NV) ஃ WahlNetwork.com ஃ @ChrisWahl ஃ Author, Networking for VMware Administrators
Reply
0 Kudos
aaron_champion
Contributor
Contributor
Jump to solution

In my case I am using 10GbE so I'm not sure I would need more than a single active port per host (I'm still working to scale this out).


My volumes span multiple physical devices (20 paths for 7 volumes per iSCSI adapter). Can I assume that with load balancing on a LAG set to hash both source and destination, I would get at least some load balancing between adapters?


If I do choose a LAG, I'm having trouble visualizing how it might reduce my iSCSI connections to my storage pool. I'm configuring the targets on my Broadcom adapters, the vmk connects through those adapters. Is the logic smart enough to only connect active paths on one adapter or another in a LAG? I'm not even sure I'm making sense anymore Smiley Wink I might have to play with a LAG on an isolated host for a bit and see how exactly it will behave.

Reply
0 Kudos
chriswahl
Virtuoso
Virtuoso
Jump to solution

My volumes span multiple physical devices (20 paths for 7 volumes per iSCSI adapter). Can I assume that with load balancing on a LAG set to hash both source and destination, I would get at least some load balancing between adapters?

If you have multiple L2-4 target endpoints, yes. Such as an IP address, MAC address, or unique ports. Paths, volumes, and devices are not relevant to LACP.

If I do choose a LAG, I'm having trouble visualizing how it might reduce my iSCSI connections to my storage pool. I'm configuring the targets on my Broadcom adapters, the vmk connects through those adapters. Is the logic smart enough to only connect active paths on one adapter or another in a LAG? I'm not even sure I'm making sense anymore Smiley Wink I might have to play with a LAG on an isolated host for a bit and see how exactly it will behave.

The adapters are no longer exposed. The LAG forms one logical interface. Traffic can be sent or received across any physical adapter in the LAG.

Note: I would not use a LAG for iSCSI traffic on a vSphere host: Seriously, Stop Using Port Channels for vSphere Storage Traffic | Wahl Network

VCDX #104 (DCV, NV) ஃ WahlNetwork.com ஃ @ChrisWahl ஃ Author, Networking for VMware Administrators
Reply
0 Kudos
aaron_champion
Contributor
Contributor
Jump to solution

Thanks, I was starting to have doubts about configuring a LAG for the iSCSI traffic earlier today when I had a chance to think about it in more detail. I can find other ways to optimize the iSCSI connection count.

Thanks for the help!

Reply
0 Kudos