VMware Cloud Community
mtsm
Contributor
Contributor

802.3ad with ip hash in ESX will increase the bandwithd??

Hi...i have a two gigabit ethernet links in my ESX configuration ....at the switch we have 802.1q and 802.3ad configured....the load balancing schema used is ip hash..

my question is....will it increase the bandwithd ? i mean....the two gigabit links will be increased into a single 2gig link? i've been reading out there..it says it will..but some friends of mine here say it wont..... what is the reality?

thnakss

0 Kudos
6 Replies
ZMkenzie
Enthusiast
Enthusiast

You don't need 802.3 ad if you use ip hash since vmware will balance machines across your links. So you will have a maximum of 1 gbit per machine with a total maximum of 2gbit per vmware server.

If you need 2gbit (and let me know which app is able to saturate 1gbit link) you have to configure 802.3ad also on vmware (instead of ip hash balancing). But this is not the best idea in my opinion.

Hope this helps.

0 Kudos
emmar
Hot Shot
Hot Shot

I was under the impression that you DO need 802.3ad when using IP hash - basically because the pswitch needs to understand that packets from the same MAC address (the VM) could appear on multiple ports.

I agree that you'd be hard pressed to find an App then saturates a 1GB and needs up to 2GB.

But i think that using Loadbalancing based on IP hash will provide the most balanced approach (although it does use more processor power than any of the other methods)

0 Kudos
Jae_Ellers
Virtuoso
Virtuoso

IOmeter is an application.

-=-=-=-=-=-=-=-=-=-=-=-=-=-=- http://blog.mr-vm.com http://www.vmprofessional.com -=-=-=-=-=-=-=-=-=-=-=-=-=-=-
0 Kudos
Paul_Lalonde
Commander
Commander

802.3ad requires the ip-hash algorithm on ESX. It's a "bidirectional" load balancing mechanism that's performed at both ends of the link. The switch would have to support it.

While it does increase the total effective throughput to 2Gbps in your case, no single flow can exceed the bandwidth of 1Gbps link. This is because flows (traffic patterns with the same src/dst) only take one of the two links, not both.

You can have TWO flows, each going across its own NIC for a total throughput of 2Gbps. But no one single flow can exceed 1Gbps.

Paul

0 Kudos
Paul_Lalonde
Commander
Commander

IP video, IP SAN, backup / restore, etc., are all apps that can easily saturate a gigabit link. With PCIe buses capable of 8GBytes/sec and faster, we'll start to see Gigabit getting saturated more and more.

Paul

0 Kudos
Andrew_Judge
Contributor
Contributor

I had a similar post regarding LACP in a 5500 3Com switch. I ended up giving up and just trunking out the ports. I've read that you can set a cisco catalyst to LACP with IP hash

port-channel load-balance src-dst-ip

But does it do any good? Probably not:( this is something that I could not do on the 3Com 5500G-EI. It just won't see the esx NIC team although it's supposedly notifying switch.

0 Kudos