DUAL 10GB for iscsi

[arrow]

Can I use Intel dual 10GB so I would have 4 x 10GB for ISCSI - tried LACP and BOND and none seems working.

 

[jgreco]

Sure, but you've not provided any context, so I could say any of {yes, no, maybe, every other Thursday} and be right in some use case (except possibly the last).

In general, the correct way to use multiple 10GbE for iSCSI is to have multiple networks. LACP only works with a large number of concurrent active iSCSI sessions due to how LACP works.

 

[arrow]

seems like I figure it out, for LACP I would need a switch not like in my case:
prox1 1x10GB Intel card
prox2 1x10GB Intel card
prox3 1x10Gb Intel Card

truenas 2x2x10GB (dual card) and have it in bond1 = all of them and I can ping whatever IP i set up for ISCSI on that bond. Indeed, I have multiple networks and one is designed for ISCSI network which I choose for target from PROX DATACENTER cluster.

 

[jgreco]

Well, that's probably not going to be doing quite what you think, and it's going to suck performance-wise compared to a proper setup, but, okay, enjoy.

 

[arrow]

I know I have to invest in switch and do it in proper way, but now for LAB it should work.

 

[jgreco]

If by work you mean (airquotes) "work", well, yeah, but you might be better off just using a single 10G on the FreeNAS side.

 

[Samuel Tai]

LACP or bond isn't the way to achieve high throughput with iSCSI, as you'll still be effectively using 10Gbps instead of 40 Gbps, as the traffic hashing will still pick only a single physical interface for egress. To achieve 40 Gbps, you'll need to set up iSCSI multipath, so each NIC is on a different subnet, and all subnets get advertised in the portal. The iSCSI initiators should also have NICs on all these subnets. When they connect to the target, they'll use all NICs.

 

[jgreco]

LACP or bond isn't the way to achieve high throughput with iSCSI, as you'll still be effectively using 10Gbps instead of 40 Gbps, as the traffic hashing will still pick only a single physical interface for egress. To achieve 40 Gbps, you'll need to set up iSCSI multipath, so each NIC is on a different subnet, and all subnets get advertised in the portal. The iSCSI initiators should also have NICs on all these subnets. When they connect to the target, they'll use all NICs.

Well, with three hosts, that's probably not true, but the likelihood of two hosts winding up on the same interface is pretty high, and one interface is guaranteed to be unused regardless.

The real problem here is that the virtual LAG interface on the NAS side is going to be imposing a significant performance penalty, and there's no benefit to the LACP. With the indicated setup, the poster would be better off doing pointopoint links for each host, using a separate subnet for each, and just forgetting all about LACP or needing a switch...