Failed to connect host to switch using LACP (IEEE 802.3ad)

[Andreev]

I have SuperMicro host VMware 6.0 with total 6 NICs and Managed Switch NetGear S3300 (24+4 port)

Want to establish NetTeaming between host and switch with protocol LAСP (IEEE 802.3ad).

At the host (https://kb.vmware.com/s/article/1004048) I establish load-balancing method: source and destination IP Hash.

At the NetGear switch I group 6 ports connected to host and change LAG type from Static to LACP.

Here the connection is lost and cannon be revived.

Please show me the way to resolve the issue.

[daphnissov]

If you're already entitled to a vDS, it's recommended (and simpler) that you not use a LAG as vSphere contains all of these abilities and then some right out of the box.

[Andreev]

vDS is not available - I have Essentials license (cannot add host to distributed switch )

The traffic is not balanced at all

vsphere-client

Monitor - Performance - Advanced - Data transmit

Object Average

vmnic0 672.5

vmnic1   0.4

vmnic2 76.0

vmnic3 583.8

vmnic4   1.3

vmnic5   0.4

switch monitor - Port statistics

port     in     out

1/g1     681M     801M

1/g2     -                52M

1/g3     3161M     3822M

1/g4     2M            54M

1/g5     -               52M

1/g6     10320M   1936M

What shall I do - Essential license is quite enough for my 3 hosts

[daphnissov]

Then an Essentials license does not entitle you to use LACP. You can only use static LAGs.

[Andreev]

Then the question is - I cannot use Static LAG at my NetGear switch as well ! ! !

I tested it at another host - Intel DZ77SL + 2 x i210 NICs

Host vSwitch Load balancing: Routing based on    <IP hash> / <source MAC hash> / <originating virtual port>

When I group 3 ports to LAG at NetGear and leave LAG type = Static -> LAG State shows <Link up> - but the host becomes offline

The question is - Can I still use Static LAG at my switch?

Another question is - When no LAG is at my switch - which Load balancing option is preferred?

I tried all three points - for 16Gb file transmit

IP hash                                            1/g1=450014          1/g2=13146740        1/g3=511151

source MAC hash                            1/g1=6872              1/g2=12418419        1/g3=18311

originating virtual port                      1/g1=11478             1/g2=12426400        1/g3=30125

like at IP hash is 1/30/1 ; at MAC is 1/2000/3 ; at VP is 1/1000/3

does this experiment indicate somethig - or not

[daphnissov]

If you only have an Essentials license, you can only use a static LAG and not LACP. If you only have 2 vmnics per ESXi host, the recommendation is do not enable a LAG because you need management to be on a non-bonded interface. If you do enable a static LAG, the ESXi host side needs to be set to route based on IP hash as this will hash src and dst IP addresses.

I tried all three points - for 16Gb file transmit

IP hash                                            1/g1=450014          1/g2=13146740        1/g3=511151

source MAC hash                            1/g1=6872              1/g2=12418419        1/g3=18311

originating virtual port                      1/g1=11478             1/g2=12426400        1/g3=30125

like at IP hash is 1/30/1 ; at MAC is 1/2000/3 ; at VP is 1/1000/3

does this experiment indicate somethig - or not

You need to read about how LAGs work because this is a very very common misconception by others. People seem to think a LAG "sprays" out data evenly across all links for all traffic and it just doesn't.

[Andreev]

The first part of updated question is YES - it works

I followed the instructions . . .

vmnic0 - onboard ; vmnic1 , vmnic2 - i210

1.vSwitch0 - Active NIC: vmnic0-1-2 ; Load balancing = originating v.p.

2.Management - Active NIC: vmnic0 / Unused vmnic1-2 ; Load balancing = originating v.p.

3.Network - Active NIC: vmnic1-2 / Unused vmnic0 ; Load balancing = IP hash (all override)

the LAG is on  - the host is on ! ! ! + + + ! ! !

Second part of updated question is disappointing . . .

16Gb to pass

NetGear LAG Static g9+g10 ; g11 = no LAG

host - g9+g10 = IP hash ; g1 = org v.p.

1/g9          1439624            0            3630

1/g10                   58           2    12414166

1/g11             13754           1          24299

this way at IP hash is 1/3400/7 = it is twice worse than MAC hash at three NICs with no LAG

I`ll try to extend the question

Shall I put m/b vmnic0 to Unused in Network - or to Standby - or even let it in Active - but not in LAG at switch ?